KR102455875B1 - Method and apparatus for bone age assessment - Google Patents

Abstract

Disclosed is a bone age reading method using a neural network performed by a computing device according to an embodiment of the present disclosure. The method includes: receiving an analysis image to be subjected to bone age reading; and inputting the analyzed image into a bone age analysis model including one or more neural networks to read bone age, and wherein the bone age analysis model is at least one for focusing on a major region of the analyzed image. However, including the above attention module (Attention Module), it may be characterized in that the supervised learning (Supervised Learning) based on the attention guide label (Attention Guide Label).

Description

Bone age reading method and device {METHOD AND APPARATUS FOR BONE AGE ASSESSMENT}

The present invention relates to a method for reading bone age, and more particularly, to a method for reading bone age using a neural network.

In the art, the Greulich-Pyle (GP) method and the Tanner-Whitehouse 3 (TW3) method are used as methods for reading the degree of bone development (bone age) from a bone image. This is also used as a ground truth for labeling in image analysis technology using artificial neural networks.

When the user who trains the artificial neural network knows the true value of the input, the supervised learning method is chosen. However, even if the true value is known, in general, artificial neural networks in which end-to-end learning is performed cannot perform more detailed supervised learning. Therefore, there has been a demand for learning to be more focused on a specific area when the true value of input is known as a whole in the field of artificial intelligence as well as in the field of artificial intelligence.

Prior paper "S. J. Son et al., "TW3-Based Fully Automated Bone Age Assessment System Using Deep Neural Networks," in IEEE Access, vol. 7, pp. 33346-33358, 2019, doi: 10.1109/ACCESS.2019.2903131." discloses an artificial neural network that has learned the TW3 method for evaluating bone age.

The present disclosure has been devised in response to the above-described background technology, and an object of the present disclosure is to provide a method for reading bone age using a neural network.

Specifically, an object of the present invention is to provide a method for supervised learning of the neural network.

A bone age reading method using a neural network performed by a computing device according to an embodiment of the present disclosure for realizing the above-described problems is disclosed. The method includes: receiving an analysis image to be subjected to bone age reading; and inputting the analyzed image into a bone age analysis model including one or more neural networks to read bone age, and wherein the bone age analysis model is at least one for focusing on a major region of the analyzed image. However, including the above attention module (Attention Module), it may be characterized in that the supervised learning (Supervised Learning) based on the attention guide label (Attention Guide Label).

In an alternative embodiment, the attention guide label is generated based on a detection result obtained as a result of inputting a training image to a main region detection model including at least one neural network, and the detection result is included in the training image. It may include location information about at least one main area.

In an alternative embodiment, when the main area is in the form of a bounding box, the detection result is the coordinates of the center point of at least one main area included in the training image, the width of the main area, and the height of the main area may include.

In an alternative embodiment, the attention guide label, in at least one pixel included in the training image, is a Gaussian distribution (Gaussian distribution) of the distance between the coordinates of the pixel and the coordinates of the center point of one or more main regions included in the training image. distribution) may include the importance of the pixel obtained as a result of substituting the equation based on the distribution).

In an alternative embodiment, the bone age analysis model, at least one or more learning images; and supervised learning based on an attention guide label corresponding to each of the learning images and including at least one main area, wherein the supervised learning includes a spatial attention map generated for the learning image using the bone age analysis model and It may be performed based on a comparison result of the attention guide label corresponding to the learning image.

In an alternative embodiment, the supervised learning may be performed based on a result calculated by substituting the spatial attention map and the attention guide label into a binary cross-entropy loss function.

In an alternative embodiment, the supervised learning of the bone age analysis model, when the bone age analysis model includes at least two or more attention modules, a predetermined attention module based on the result value of the loss function calculated in each attention module It may be performed based on the summed result after multiplying the weights.

In an alternative embodiment, the attention module may include: a channel attention neural network model for generating a channel attention map with respect to the feature map input to the attention module; and a spatial attention neural network model for generating a spatial attention map with respect to the modified feature map, wherein the modified feature map is a feature map in which the channel attention map is multiplied by element by the feature map input to the attention module. have.

A bone age reading method using a neural network performed by a computing device according to an embodiment of the present disclosure for realizing the above-described problems is disclosed. The method includes: receiving an analysis image to be subjected to bone age reading; Reading the bone age by inputting the analyzed image into a bone age analysis model including one or more neural networks and providing a user interface screen including a heat map generated based on a spatial attention map for the analyzed image; Including, and the bone age analysis model, including at least one or more attention module (Attention Module) for centrally analyzing the main area of the analyzed image, guided learning based on the attention guide label (Attention Guide Label) (Supervised Learning) may be characterized.

A bone age reading method using a neural network performed by a computing device according to an embodiment of the present disclosure for realizing the above-described problems is disclosed. The method includes: receiving an analysis image to be subjected to bone age reading; Reading the bone age by inputting the analyzed image into a bone age analysis model including one or more neural networks and providing a user interface screen including a heat map generated based on a spatial attention map for the analyzed image; Including, and the attention module (Attention Module) included in the bone age analysis model, a channel attention neural network model for generating a channel attention map with respect to the feature map input to the attention module; and a spatial attention neural network model for generating a spatial attention map with respect to the modified feature map, wherein the modified feature map may be a feature map in which the channel attention map is multiplied by element by the feature map input to the attention module. have.

Disclosed is a computer program stored in a computer-readable storage medium according to an embodiment of the present disclosure for realizing the above-described problems. The computer program, when executed in one or more processors, performs the following operations for reading bone age using a neural network, wherein the operations include: Including the operation of reading the bone age by input to the bone age analysis model, and the bone age analysis model, including at least one attention module (Attention Module) for centrally analyzing the main area of the analyzed image, It may be characterized in that it is supervised learning based on the attention guide label (Attention Guide Label).

Disclosed is an apparatus for reading bone age according to an embodiment of the present disclosure for realizing the above-described problem. The apparatus may include one or more processors; and a memory storing a bone age analysis model comprising one or more neural networks; A network unit for receiving an analysis image that is a target of bone age reading, and the one or more processors input the analyzed image to a bone age analysis model including one or more neural networks to read the age of the bone, and the bone age The age analysis model includes at least one attention module for centrally analyzing the main area of the analyzed image, but is supervised learning based on the attention guide label can be done with

The present disclosure may provide a method for reading bone age using a neural network.

1 is a block diagram of a computing device for reading bone age according to an embodiment of the present disclosure.
2 is a schematic diagram illustrating a network function according to an embodiment of the present disclosure;
3 is an exemplary diagram illustrating a structure of an attention module according to an embodiment of the present disclosure.
4 is an exemplary view illustrating an attention guide label according to an embodiment of the present disclosure.
5 is an exemplary diagram illustrating a result of using a main region detection model according to an embodiment of the present disclosure.
6 is an exemplary diagram illustrating a user interface screen including a heat map generated based on a spatial attention map for an analysis image according to an embodiment of the present disclosure.
7 is a flowchart for a computing device to read a bone age for an analyzed image and provide a user interface screen according to an embodiment of the present disclosure.
8 is a simplified, general schematic diagram of an exemplary computing environment in which embodiments of the present disclosure may be implemented.

Various embodiments are now described with reference to the drawings. In this specification, various descriptions are presented to provide an understanding of the present disclosure. However, it is apparent that these embodiments may be practiced without these specific descriptions.

The terms “component,” “module,” “system,” and the like, as used herein, refer to a computer-related entity, hardware, firmware, software, a combination of software and hardware, or execution of software. For example, a component can be, but is not limited to being, a process running on a processor, a processor, an object, a thread of execution, a program, and/or a computer. For example, both an application running on a computing device and the computing device may be a component. One or more components may reside within a processor and/or thread of execution. A component may be localized within one computer. A component may be distributed between two or more computers. In addition, these components can execute from various computer readable media having various data structures stored therein. Components may communicate via a network such as the Internet with another system, for example via a signal having one or more data packets (eg, data and/or signals from one component interacting with another component in a local system, distributed system, etc.) may communicate via local and/or remote processes depending on the data being transmitted).

In addition, the term “or” is intended to mean an inclusive “or” rather than an exclusive “or.” That is, unless otherwise specified or clear from context, "X employs A or B" is intended to mean one of the natural implicit substitutions. That is, X employs A; X employs B; or when X employs both A and B, "X employs A or B" may apply to either of these cases. It should also be understood that the term “and/or” as used herein refers to and includes all possible combinations of one or more of the listed related items.

Also, the terms "comprises" and/or "comprising" should be understood to mean that the feature and/or element in question is present. However, it should be understood that the terms "comprises" and/or "comprising" do not exclude the presence or addition of one or more other features, elements and/or groups thereof. Also, unless otherwise specified or unless it is clear from context to refer to a singular form, the singular in the specification and claims should generally be construed to mean “one or more”.

In addition, the term “at least one of A or B” should be interpreted to mean “includes only A”, “includes only B”, and “combined with the configuration of A and B”.

Those skilled in the art will further appreciate that the various illustrative logical blocks, configurations, modules, circuits, means, logics, and algorithm steps described in connection with the embodiments disclosed herein may be implemented in electronic hardware, computer software, or combinations of both. It should be recognized that they can be implemented with To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, configurations, means, logics, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application. However, such implementation decisions should not be interpreted as causing a departure from the scope of the present disclosure.

Descriptions of the presented embodiments are provided to enable those skilled in the art to use or practice the present invention. Various modifications to these embodiments will be apparent to those skilled in the art of the present disclosure. The generic principles defined herein may be applied to other embodiments without departing from the scope of the present disclosure. Thus, the present invention is not limited to the embodiments presented herein. The invention is to be construed in its widest scope consistent with the principles and novel features presented herein.

The term "image" or "image data" as used throughout the present description and claims refers to multidimensional data composed of discrete image elements (eg, pixels in a two-dimensional image), in other words, ( For example, a term that refers to a visible object (eg, displayed on a video screen) or a digital representation of the object (eg, a file corresponding to a pixel output of a CT, MRI detector, etc.).

For example, in the present disclosure, "image" or "image" means computed tomography (CT), magnetic resonance imaging (MRI), fundus imaging, ultrasound, or any known in the art. It may mean a medical image of a subject collected by another medical imaging system of The image does not necessarily have to be provided in a medical context, but may be provided in a non-medical context, for example, there may be X-ray imaging for security screening.

Throughout the detailed description and claims of the present invention, the 'DICOM (Digital Imaging and Communications in Medicine)' standard is a generic term for various standards used for digital image expression and communication in medical devices. The DICOM standard is published by a joint committee formed by the American Society of Radiological Medicine (ACR) and the National Electrical Engineers Association (NEMA).

Throughout the detailed description and claims of the present invention, 'Picture Archiving and Communication System (PACS)' is a term that refers to a system that stores, processes, and transmits in accordance with the DICOM standard, and includes X-ray, CT, MRI Medical image images acquired using digital medical imaging equipment such as DICOM are stored in DICOM format and can be transmitted to terminals inside and outside the hospital through a network, to which reading results and medical records can be added.

1 is a block diagram of a computing device for reading bone age according to an embodiment of the present disclosure.

The configuration of the computing device 100 shown in FIG. 1 is only a simplified example. In an embodiment of the present disclosure, the computing device 100 may include other components for performing the computing environment of the computing device 100 , and only some of the disclosed components may configure the computing device 100 .

The computing device 100 may include a processor 110 , a memory 130 , and a network unit 150 .

The processor 110 may include one or more cores, and a central processing unit (CPU) of a computing device, a general purpose graphics processing unit (GPGPU), and a tensor processing unit (TPU). unit) and the like, and may include a processor for data analysis and deep learning. The processor 110 may read a computer program stored in the memory 130 to perform data processing for machine learning according to an embodiment of the present disclosure. According to an embodiment of the present disclosure, the processor 110 may perform an operation for learning the neural network. The processor 110 for learning of the neural network, such as processing input data for learning in deep learning (DL), extracting features from the input data, calculating an error, updating the weight of the neural network using backpropagation calculations can be performed. At least one of a CPU, a GPGPU, and a TPU of the processor 110 may process learning of a network function. For example, the CPU and the GPGPU can process learning of a network function and data classification using the network function. Also, in an embodiment of the present disclosure, learning of a network function and data classification using the network function may be processed by using the processors of a plurality of computing devices together. In addition, the computer program executed in the computing device according to an embodiment of the present disclosure may be a CPU, GPGPU or TPU executable program.

According to an embodiment of the present disclosure, the memory 130 may store any type of information generated or determined by the processor 110 and any type of information received by the network unit 150 .

According to an embodiment of the present disclosure, the memory 130 is a flash memory type, a hard disk type, a multimedia card micro type, and a card type memory (eg, For example, SD or XD memory), Random Access Memory (RAM), Static Random Access Memory (SRAM), Read-Only Memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Programmable Read (PROM) -Only Memory), a magnetic memory, a magnetic disk, and an optical disk may include at least one type of storage medium. The computing device 100 may operate in relation to a web storage that performs a storage function of the memory 130 on the Internet. The description of the above-described memory is only an example, and the present disclosure is not limited thereto.

The network unit 150 according to an embodiment of the present disclosure includes a Public Switched Telephone Network (PSTN), x Digital Subscriber Line (xDSL), Rate Adaptive DSL (RADSL), Multi Rate DSL (MDSL), VDSL ( A variety of wired communication systems such as Very High Speed DSL), Universal Asymmetric DSL (UADSL), High Bit Rate DSL (HDSL), and Local Area Network (LAN) can be used.

In addition, the network unit 150 presented herein is CDMA (Code Division Multi Access), TDMA (Time Division Multi Access), FDMA (Frequency Division Multi Access), OFDMA (Orthogonal Frequency Division Multi Access), SC-FDMA ( A variety of wireless communication systems may be used, such as Single Carrier-FDMA) and other systems.

In the present disclosure, the network unit 150 may be configured regardless of its communication mode, such as wired and wireless, and may be configured with various communication networks such as a personal area network (PAN) and a wide area network (WAN). can In addition, the network may be a well-known World Wide Web (WWW), and may use a wireless transmission technology used for short-range communication such as Infrared Data Association (IrDA) or Bluetooth.

The techniques described herein may be used in the networks mentioned above, as well as in other networks.

A bone age reading method using a neural network performed by the computing device of the present disclosure includes: receiving an analysis image that is a target of bone age reading; and inputting the analyzed image into a bone age analysis model including one or more neural networks to read the bone age. And the bone age analysis model, including at least one attention module (Attention Module) for centrally analyzing the main area of the analyzed image, is supervised learning based on the attention guide label (Attention Guide Label) can be characterized.

In an embodiment of the present disclosure, the network unit 150 included in the computing device 100 may receive an analysis image that is a target of bone age reading. The bone age reading means measuring or predicting bone maturity from a bone-related medical image. The analyzed image may be a bone-related image that requires reading. In the present specification, 'image' may be used as a concept including all kinds of medical images that can be input to the bone age reading model, including analysis images or learning images. Therefore, in the present disclosure, the analysis image or the learning image that is the target of bone age reading may include, for example, at least one of an X-ray image, a CT image, and an MRI image related to the bone provided for determining the age of the bone. . In addition, the analysis image or the learning image may include, without limitation, any bone-related image that can be a target of bone age reading, such as a hand bone, an elbow bone, a knee bone, or the like. The training image is an image used for learning the bone age reading model, and may be distinguished from an analysis image that is an image input in the inference step of the bone age reading model.

In an embodiment of the present disclosure, the processor 110 included in the computing device 100 may read the bone age by inputting the analyzed image into a bone age analysis model including one or more neural networks.

2 is a schematic diagram illustrating a network function according to an embodiment of the present disclosure;

Throughout this specification, computational model, neural network, network function, and neural network may be used interchangeably. A neural network may be composed of a set of interconnected computational units, which may generally be referred to as nodes. These nodes may also be referred to as neurons. A neural network is configured to include at least one or more nodes. Nodes (or neurons) constituting the neural networks may be interconnected by one or more links.

In the neural network, one or more nodes connected through a link may relatively form a relationship between an input node and an output node. The concepts of an input node and an output node are relative, and any node in an output node relationship with respect to one node may be in an input node relationship in a relationship with another node, and vice versa. As described above, an input node-to-output node relationship may be created around a link. One or more output nodes may be connected to one input node through a link, and vice versa.

In the relationship between the input node and the output node connected through one link, the value of the data of the output node may be determined based on data input to the input node. Here, a link interconnecting the input node and the output node may have a weight. The weight may be variable, and may be changed by the user or algorithm in order for the neural network to perform a desired function. For example, when one or more input nodes are interconnected to one output node by respective links, the output node sets values input to input nodes connected to the output node and links corresponding to the respective input nodes. An output node value may be determined based on the weight.

As described above, in a neural network, one or more nodes are interconnected through one or more links to form an input node and an output node relationship in the neural network. The characteristics of the neural network may be determined according to the number of nodes and links in the neural network, the correlation between the nodes and the links, and the value of a weight assigned to each of the links. For example, when the same number of nodes and links exist and there are two neural networks having different weight values of the links, the two neural networks may be recognized as different from each other.

A neural network may consist of a set of one or more nodes. A subset of nodes constituting the neural network may constitute a layer. Some of the nodes constituting the neural network may configure one layer based on distances from the initial input node. For example, a set of nodes having a distance n from the initial input node may constitute n layers. The distance from the initial input node may be defined by the minimum number of links that must be traversed to reach the corresponding node from the initial input node. However, the definition of such a layer is arbitrary for description, and the order of the layer in the neural network may be defined in a different way from the above. For example, a layer of nodes may be defined by a distance from the final output node.

The initial input node may mean one or more nodes to which data is directly input without going through a link in a relationship with other nodes among nodes in the neural network. Alternatively, in a relationship between nodes based on a link in a neural network, it may mean nodes that do not have other input nodes connected by a link. Similarly, the final output node may refer to one or more nodes that do not have an output node in relation to other nodes among nodes in the neural network. In addition, the hidden node may mean nodes constituting the neural network other than the first input node and the last output node.

The neural network according to an embodiment of the present disclosure may be a neural network in which the number of nodes in the input layer may be the same as the number of nodes in the output layer, and the number of nodes decreases and then increases again as the input layer progresses to the hidden layer. can In addition, the neural network according to another embodiment of the present disclosure may be a neural network in which the number of nodes in the input layer may be less than the number of nodes in the output layer, and the number of nodes decreases as the number of nodes progresses from the input layer to the hidden layer. have. In addition, the neural network according to another embodiment of the present disclosure may be a neural network in which the number of nodes in the input layer may be greater than the number of nodes in the output layer, and the number of nodes increases as the number of nodes progresses from the input layer to the hidden layer. can The neural network according to another embodiment of the present disclosure may be a neural network in a combined form of the aforementioned neural networks.

A deep neural network (DNN) may refer to a neural network including a plurality of hidden layers in addition to an input layer and an output layer. Deep neural networks can be used to identify the latent structures of data. In other words, it can identify the potential structure of photos, texts, videos, voices, and music (e.g., what objects are in the photos, what the text and emotions are, what the texts and emotions are, etc.) . Deep neural networks include convolutional neural networks (CNNs), recurrent neural networks (RNNs), auto encoders, generative adversarial networks (GANs), and restricted boltzmann machines (RBMs). machine), a deep belief network (DBN), a Q network, a U network, a Siamese network, and a Generative Adversarial Network (GAN). The description of the deep neural network described above is only an example, and the present disclosure is not limited thereto.

In an embodiment of the present disclosure, the network function may include an autoencoder. The auto-encoder may be a kind of artificial neural network for outputting output data similar to input data. The auto encoder may include at least one hidden layer, and an odd number of hidden layers may be disposed between the input/output layers. The number of nodes in each layer may be reduced from the number of nodes in the input layer to an intermediate layer called the bottleneck layer (encoding), and then expanded symmetrically with reduction from the bottleneck layer to the output layer (symmetrical to the input layer). Autoencoders can perform non-linear dimensionality reduction. The number of input layers and output layers may correspond to a dimension after preprocessing the input data. In the auto-encoder structure, the number of nodes of the hidden layer included in the encoder may have a structure that decreases as the distance from the input layer increases. If the number of nodes in the bottleneck layer (the layer with the fewest nodes between the encoder and decoder) is too small, a sufficient amount of information may not be conveyed, so a certain number or more (e.g., more than half of the input layer, etc.) ) may be maintained.

The neural network may be trained using at least one of supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning. Learning of the neural network may be a process of applying knowledge for the neural network to perform a specific operation to the neural network.

A neural network can be trained in a way that minimizes output errors. In the training of a neural network, iteratively input the training data into the neural network, calculate the output of the neural network and the target error for the training data, and calculate the error of the neural network from the output layer of the neural network to the input layer in the direction to reduce the error. It is the process of updating the weight of each node of the neural network by backpropagation in the direction. In the case of teacher learning, learning data in which the correct answer is labeled in each learning data is used (ie, labeled learning data), and in the case of comparative learning, the correct answer may not be labeled in each learning data. That is, for example, the learning data in the case of teacher learning regarding data classification may be data in which categories are labeled for each of the learning data. Labeled training data is input to the neural network, and an error can be calculated by comparing the output (category) of the neural network with the label of the training data. As another example, in the case of comparison learning about data classification, an error may be calculated by comparing the input training data with the output of the neural network. The calculated error is back propagated in the reverse direction (ie, from the output layer to the input layer) in the neural network, and the connection weight of each node of each layer of the neural network may be updated according to the back propagation. A change amount of the connection weight of each node to be updated may be determined according to a learning rate. The computation of the neural network on the input data and the backpropagation of errors can constitute a learning cycle (epoch). The learning rate may be applied differently depending on the number of repetitions of the learning cycle of the neural network. For example, in the early stage of learning of a neural network, a high learning rate can be used to enable the neural network to quickly acquire a certain level of performance, thereby increasing efficiency, and using a low learning rate at the end of learning can increase accuracy.

In training of a neural network, in general, the training data may be a subset of real data (that is, data to be processed using the trained neural network), and thus the error on the training data is reduced, but the error on the real data is reduced. There may be increasing learning cycles. Overfitting is a phenomenon in which errors on actual data increase by over-learning on training data as described above. For example, a phenomenon in which a neural network that has learned a cat by seeing a yellow cat does not recognize that it is a cat when it sees a cat other than yellow may be a type of overfitting. Overfitting can act as a cause of increasing errors in machine learning algorithms. In order to prevent such overfitting, various optimization methods can be used. In order to prevent overfitting, methods such as increasing the training data, regularization, and dropout that deactivate some of the nodes of the network in the process of learning, and the use of a batch normalization layer are applied. can

Hereinafter, the structure of the bone age analysis model including the one or more neural networks will be described.

In an embodiment of the present disclosure, the bone age analysis model may include at least one attention module for centrally analyzing a main region of the analyzed image. The main region is a concept meaning a portion of an image to be considered as a priority in a process in which the processor 110 calculates a final output from an image through a bone age analysis model. The image may include a learning image used for learning the bone age analysis model or an analysis image used for inference through the bone age analysis model.

In an embodiment of the present disclosure, the main region may include a partial region related to at least one hand joint to be checked in the Tanner-Whitehouse 3 (TW3) method of reading bone age. In another embodiment, the main region may include at least one partial region according to the TW2 method, at least one partial region required to be confirmed in order to estimate the bone age from the elbow joint image, and the like. The above-described example of the main region is only an example, and the present disclosure includes, without limitation, the main region that the analysis model should focus on in order to read the bone age from the bone-related image. In an embodiment of the present disclosure, the main area may include pixel coordinates, a width, a height, and the like.

In the attention module included in the bone age analysis model according to an embodiment of the present disclosure, at least one block or node that is connected to the attention module and constitutes the next layer of the bone age analysis model is a main region The internal value of the feature map related to the main region may be adjusted so that the internal value of the feature map related to the main region is significantly influenced by the internal value of the feature map not related to the main region. The feature map may include at least one of an intermediate output feature map of several neural network layers included in the bone age analysis model during a calculation process for an analysis image or a learning image input to the bone age analysis model. The feature map may have a three-dimensional size as a result of, for example, generating a two-dimensional array as many as the number of channels according to the type of filter. Alternatively, the feature map may have an arbitrary N-dimensional size. The internal value of the feature map may mean data included in a multidimensional array. The data may be represented as a real number. The adjustment of the internal value of the feature map related to the main area of the attention module is an operation for amplifying the absolute value of the internal value of the relevant feature map through appropriate weight learning, or an operation for reducing the absolute value of the internal value of the unrelated feature map. and the like.

The attention module according to an embodiment of the present disclosure includes a channel attention neural network model for generating a channel attention map with respect to a feature map input to the attention module, and a spatial attention neural network for generating a spatial attention map with respect to the modified feature map. Models can be included. The modified feature map may be a feature map in which the channel attention map is multiplied by element by the feature map input to the attention module. The attention module may exist at any location between blocks other than the attention module included in the bone age analysis model, and at least one may exist. In the attention module, the channel attention neural network model and the spatial attention neural network model may maintain a serial connection structure in which, for example, a channel attention neural network model precedes and a spatial attention neural network model follows.

A channel attention neural network model according to an embodiment of the present disclosure applies a pooling technique to a two-dimensional array corresponding to each channel along a channel axis of the feature map in a feature map input to a three-dimensional attention module, Based on this, a channel attention map may be generated. The pooling technique may include a global max pooling technique or a global average pooling technique. Specifically, the channel attention neural network model may input the feature map to which the pooling technique is applied to a first network function including at least one connection weight or bias, and generate a channel attention map as a result. Two or more feature maps to which the pooling technique is applied may be generated according to the type of the pooling technique.

For example, suppose that the feature map input to the attention module is a multidimensional array having a height of 64, a width of 64, and a size of a channel C. For this 64x64xC size feature map, the processor 110 may generate a 1x1xC size channel attention intermediate feature map by applying a pooling technique along the channel axis through the channel attention neural network model. The channel attention intermediate feature map may be a substantially one-dimensional array. The channel attention intermediate feature map may include, for example, a first channel attention intermediate feature map to which a global average pooling technique is applied and a second channel attention intermediate feature map to which a global maximum pooling technique is applied. The channel attention neural network model may input at least one of the channel attention intermediate feature maps to the first network function and generate a channel attention map as a result. The first network function may include, for example, a multi-layer perceptron including at least one hidden layer. When the first network function consists of a multi-layer perceptron, the number of nodes of the hidden layer may be less than the size of the channel attention intermediate feature map by a certain percentage or less in order to reduce the number of connection weights and bias values and increase the operation speed. At least one connection weight or bias value included in the first network function may be supervised learning by a label related to a channel attention map, and semi-supervised learning is performed in the training process of a spatial attention neural network model, which will be described later. it might be The example regarding the size of the array or the type of the first network function is only an example for description and does not limit the present disclosure.

The spatial attention neural network model according to an embodiment of the present disclosure may generate a spatial attention map with respect to the modified feature map. The modified feature map means a feature map generated as a result of applying the channel attention map generated by the channel attention neural network model to the feature map input to the attention module. The modified feature map may be generated as a result of element-wise multiplication of the feature map input to the attention module and the channel attention map. The spatial attention neural network model of the present disclosure generates a spatial attention map with respect to the modified feature map, so that the attention module of the present disclosure can obtain an effect of arranging the channel attention neural network model and the spatial attention neural network model in a predetermined order.

In the present disclosure, the multiplication operation for each element may include an operation for multiplying elements at the same position when the two multidimensional arrays have the same size. The multiplication operation for each element may include an operation of multiplying the elements of the same position after making the array sizes the same through a broadcasting technique when the sizes of the two multidimensional arrays are different from each other. The broadcasting technique refers to a technique of duplicating one array to have the same size as the other array when the sizes of the two arrays are different from each other. For example, if the first array to be multiplied by element has a size of 64x64xC and the second array has a size of 1x1xC, the two arrays have the same number of channels of the size 'C', but the height and width are ' Since the second array is 64' times different, an array having a size of 64x64xC can be created by duplicating the second array and concatenating it in the height and width directions, and then multiplying the array with the first array element by element. For another example, if the first array to be multiplied by element has a size of 64x64xC and the second array has a size of 64x64x1, the two arrays have the same height and width of '64', but the number of channels Since it is different by 'C' times, an array having a size of 64x64xC can be created by duplicating the second array and concatenating it in the channel direction, and then multiplying the array with the first array element by element. The example of the multiplication operation for each element is only an example and does not limit the present disclosure.

The spatial attention neural network model according to an embodiment of the present disclosure is 1 corresponding to each (h,w) coordinate along the height (H) axis or the width (W) axis perpendicular to the channel axis in the three-dimensional modified feature map. A pooling technique may be applied to a dimensional array and a spatial attention map may be generated based on the pooling technique. The (h,w) coordinates refer to coordinates on a two-dimensional plane including the remaining height and width axes except for the channel axis in the three-dimensional feature map. The one-dimensional array corresponding to the (h,w) coordinate means an array including data corresponding to each of the entire channels. The pooling technique may include the above-described global maximum pooling technique or the global average pooling technique.

For example, it is assumed that the feature map modified by applying the channel attention map is a multidimensional array having a height of 64, a width of 64, and a size of channel C. In this case, the processor 110 applies the pooling technique to the entire channel corresponding to the (h,w) coordinates along the height (H) axis or the width (W) axis of the modified feature map through the spatial attention neural network model to 64x64x1 A spatial attention medium feature map of the size can be generated. Two or more spatial attention intermediate feature maps may be generated according to a type of a pooling technique. The spatial attention intermediate feature map may include, for example, a first spatial attention intermediate feature map based on a global average pooling technique or a second spatial attention intermediate feature map based on a global maximum pooling technique. The spatial attention neural network model may input the spatial attention intermediate feature map to a second network function and generate a spatial attention map as a result. The second network function may include, for example, a convolutional artificial neural network (CNN) including at least one node. The convolutional artificial neural network of the second network function may be supervised, and the learning method will be described later in detail. The spatial attention neural network model may generate a spatial attention map by inputting at least one or more spatial attention intermediate feature maps as described above to the second network function.

Hereinafter, the attention module will be further described with reference to FIG. 3 . 3 is an exemplary diagram illustrating a structure of an attention module according to an embodiment of the present disclosure. According to an embodiment of the present disclosure, the processor 110 generates the channel attention map 311 after inputting the feature map 310 input to the attention module to the channel attention neural network model. Thereafter, the processor 110 generates a modified feature map 330 by element-wise multiplication 313 of the channel attention map 311 and the feature map 310 input to the attention module. At this time, since the channel attention map 311 has half the height and width of the feature map 310 input to the attention module, the processor 110 performs elementwise multiplication 313 after broadcasting as described above. . The processor 110 generates the spatial attention map 331 after inputting the modified feature map 330 into the spatial attention neural network model. Thereafter, the processor 110 may generate an output feature map 350 of the attention module by element-wise multiplication 313 of the generated spatial attention map 331 and the modified feature map 330 . In the output feature map 350 of the attention module, when compared with the feature map 310 input to the attention module, the values of parameters related to the main region are amplified and the values of parameters irrelevant to the main region are offset. The above-described example is merely an example and does not limit the present disclosure.

As described above, the attention module according to an embodiment of the present disclosure may include a channel attention neural network model for generating a channel attention map and a spatial attention neural network model for generating a spatial attention map. The channel attention map may be generated with respect to the feature map input to the attention module. The spatial attention map may be generated for the modified feature map. The modified feature map may be generated by the processor 110 multiplying the channel attention map element by element with respect to the feature map input to the attention module. The processor 110 may generate an output feature map of the attention module by multiplying the modified feature map by the spatial attention map for each element. The output feature map of the attention module is a feature map in which a channel attention map and a spatial attention map are sequentially applied to the feature map input to the attention module. have

In an embodiment of the present disclosure, the bone age analysis model may be supervised based on the attention guide label. In the present disclosure, the attention guide label may mean data having an importance value for each at least one pixel of the training image. The importance may be expressed as a discrete distribution or a continuous distribution. For example, when the importance is expressed as a discrete distribution, the importance may be expressed as a binary discrete distribution including 0 and 1 or a multidimensional discrete distribution including two or more differential values. For another example, when the importance is expressed as a continuous distribution, the importance may be expressed as a probability value within a predetermined interval. The probability value may include a normally distributed probability value.

In one embodiment of the present disclosure, the attention guide label is a Gaussian distribution (Gaussian distribution) of the distance between the coordinates of the pixel and the coordinates of the center point of one or more main regions included in the training image in at least one pixel included in the training image. distribution) may include the importance of the pixel obtained as a result of substituting the equation based on the distribution). The importance of the pixel may exist for each pixel for each pixel. The distance may include, for example, a Euclidean distance, a Manhattan distance, or a Chebyshev distance. As an embodiment, when the processor 110 calculates the importance of each pixel based on the Euclidean distance, an equation for calculating the distance may be expressed as Equation (1).

[Equation 1]

는 거리 함수를 나타낸다.

는 학습영상에 포함된 픽셀의 좌표를 나타낸다.

는 상기 적어도 하나 이상의 주요 영역이 총

개 존재할 경우,

번째 주요 영역을 나타내는 기호이며,

및

는

번째 주요 영역의 중심점의

좌표 및

좌표를 순서대로 나타낸다. 프로세서(110)는 상기 수학식 1에 따라 계산된 거리 값을 가우시안 분포에 기초한 계산식에 대입하여 각 픽셀의 중요도를 산출할 수 있다. 상기 가우시안 분포에 기초한 계산식은 수학식 2와 같이 표현될 수 있다.in Equation 1

represents the distance function.

represents the coordinates of pixels included in the training image.

is the total of the at least one or more major areas

dog, if present,

It is a symbol indicating the second main area,

and

Is

of the central point of the second major region

coordinates and

The coordinates are displayed in order. The processor 110 may calculate the importance of each pixel by substituting the distance value calculated according to Equation 1 into a calculation formula based on a Gaussian distribution. A calculation formula based on the Gaussian distribution may be expressed as Equation (2).

[Equation 2]

는 확률 함수를 나타낸다.

는 학습영상에 포함된 각 픽셀의 좌표를 나타낸다.

는 상기 적어도 하나 이상의 주요 영역이 총

개 존재할 경우,

번째 주요 영역을 나타내는 기호이며,

는 전체 주요 영역의 집합을 나타낸다. 수학식 2와 같이 표현되는 가우시안 분포에 기초한 계산식은 픽셀의 중요도를 확률 값으로 가진다.

는

번째 주요 영역인

에 대한 분산을 나타낸다. 분산을 나타내는

함수는 수학식 3과 같이 표현될 수 있다.in Equation 2

represents a probability function.

represents the coordinates of each pixel included in the training image.

is the total of the at least one or more major areas

dog, if present,

It is a symbol indicating the second main area,

represents the set of all major domains. A calculation formula based on a Gaussian distribution expressed as in Equation 2 has the importance of a pixel as a probability value.

Is

The second major area

represents the variance for representing the variance

The function can be expressed as Equation (3).

[Equation 3]

는

번째 주요 영역인

가 직사각형으로 표시될 경우 각각 그 너비 및 높이를 나타낸다.

는 분산 계산을 위한 사전 결정된 값을 나타낸다.

Is

The second major area

When is displayed as a rectangle, it indicates the width and height, respectively.

denotes a predetermined value for variance calculation.

As described above, for one learning image, if the coordinates of the center point of at least one main region, the width of the main region, and the height of the main region exist, all of the values included in the learning image using Equations 1 to 3 are present. For a pixel, the importance of a pixel can be calculated. The method for calculating a function or probability related to the above-described distance is merely an example and does not limit the present disclosure.

4 is an exemplary view illustrating an attention guide label according to an embodiment of the present disclosure. When the main region in the bone-related learning image 410 is a region near at least one or more joint nodes, the processor 110 controls all pixels included in the bone-related learning image 410 based on Equation 1 above. Areas and distances can be calculated. Thereafter, the processor 110 may calculate a variance for the main region based on Equation 3 above. A constant for calculating variance may be set to 8. Thereafter, the processor 110 may substitute the results of Equations 1 and 3 into Equation 2 to finally calculate the importance of all pixels included in the bone-related learning image 410 . Reference numeral 430 of FIG. 4 is an exemplary diagram in which a pixel having a higher importance is displayed in a darker color, and a pixel having a lower importance is displayed in a brighter color. The processor 110 may generate an attention guide label corresponding to the training image in the same manner as described above.

In an embodiment of the present disclosure, the attention guide label may be generated based on a detection result obtained as a result of the processor 110 inputting a learning image into a main region detection model including at least one or more neural networks. The detection result may include location information on at least one main area included in the training image. The learning image may be an image for learning the bone age analysis model of the present disclosure. The main region detection model may be a model for detecting and outputting one or more main regions from an input image by having a convolutional artificial neural network structure. At least one connection weight or bias value included in the main region detection model may be learned through separate learning by the processor 110 . At least one connection weight or bias value included in the main region detection model may be transmitted through the network unit 150 after external learning is completed and stored in the memory 130 . As an embodiment, when learning the main region detection model as a learning image related to the bone, the main region detection model detects the test target region by the TW3 method from the training image related to the bone as the main region, and provides location information about the main region. It is possible to output a detection result including The detection result may be data including position information on a main area included in the learning image as a numerical value for each learning image. As an embodiment, when the main area detected by the main area detection model has a rectangular shape, the location information on the main area may include coordinates of four vertices. As another embodiment, when the main area detected by the main area detection model has a rectangular shape, location information on the main area may include a midpoint, a width, and a height of the rectangle. As another embodiment, when the main area detected by the main area detection model has a circular shape, the location information on the main area may include a midpoint and a radius of the main area. The example regarding the location information is only an example and does not limit the present disclosure.

In an embodiment of the present disclosure, when the main area is in the form of a bounding box, the detection result using the main area detection model of the processor 110 is the coordinates of the center point of at least one main area included in the training image, It may include the width of the main area and the height of the main area. The coordinates of the central point of the main region may be the midpoints of the coordinates of the four corners of the bounding box.

5 is an exemplary diagram illustrating a result of using a main region detection model according to an embodiment of the present disclosure. The processor 110 may learn the main region detection model to detect the main region 511 from the bone image 510 after inputting the bone image 510 for learning the main region detection model to the main region detection model. . The trained main region detection model detects at least one main region 511 with respect to an input image, and coordinates the center point of the main region 511 , the width of the main region 511 , and the height of the main region 511 . and the like.

In an embodiment of the present disclosure, the attention guide label may be generated based on the display result of the main region included in the user's learning image. The user may generate an attention guide label corresponding to the learning image by displaying one or more main regions for the learning image and providing them to the processor 110 as a result of displaying the main regions. Specifically, the user may display on the training image at least one or more major areas determined to be analyzed intensively by the bone age analysis model in the learning image. The operation of the user to display the main area on the learning image may include, for example, an operation of inputting at least one of the coordinates of the center point, the width of the main area, or the height of the main area when the main area is in the form of a bounding box. can The processor 110 may recognize the location information of the corresponding area from the display result of the user's main area. For example, when the user displays the main area as a rectangle, the processor 110 may recognize the coordinates, width, and height of the center point of the main area. An operation in which the user displays the main area on the learning image may be performed based on a separate image viewer program or the like. The processor 110 may generate an attention guide label through Equations 1 to 3 described above using the learning image and the display result of the user's main region corresponding to the learning image.

In another embodiment of the present disclosure, the user may directly generate an attention guide label for the learning image. That is, the user may directly perform a labeling operation of generating an attention guide label corresponding to the learning image for the learning image. For example, the user may directly input the importance of at least one pixel included in the learning image for learning into the processor 110 . When the user performs the labeling operation on the training image, the processor 110 may set the remaining pixels except for one or more pixels to which the user directly inputs the importance of the pixel to have a predetermined importance. The predetermined importance may be set to zero, for example. The processor 110 may match the attention guide label labeled by the user with the corresponding learning image to use it for learning the bone age analysis model.

Hereinafter, the supervised learning method of the bone age analysis model described above will be described. In an embodiment of the present disclosure, the bone age analysis model is supervised learning based on at least one or more learning images and an attention guide label corresponding to each of the learning images and including at least one or more main areas, and the supervised learning is the goal This may be performed based on a comparison result between a spatial attention map generated for the learning image using an age analysis model and an attention guide label corresponding to the learning image. In the present disclosure, the bone age analysis model may use at least one loss value for supervised learning of the model. The at least one loss value is, for example, a regression loss by comparing a bone age prediction value with an actual age or a regression loss by comparing an attention guide label with a spatial attention map generated by the bone age analysis model in the calculation process of the training image. may include The processor 110 may update the internal connection weight or bias value in a direction in which the loss value decreases during the learning process of the bone age analysis model. The processor 110 may use, for example, at least one of a root mean square error function, a mean square error function, a mean absolute value error function, and a mean square log error function as a loss function for calculating the regression loss. In addition, the processor 110 may use a cross entropy function or a binary cross entropy function as a loss function for calculating the regression loss.

As described above, for supervised learning of the bone age analysis model, the processor 110 may compare the spatial attention map generated by the bone age analysis model for the training image with the attention guide label for the training image. The comparison includes comparing data corresponding to elements at the same position in an array of the same size. For the comparison, when the size of the spatial attention map is smaller than the size of the attention guide label, the processor 110 may perform upsampling on the spatial attention map. The upsampling may include unpooling. The upsampling may refer to an operation of increasing the height and width sizes of the spatial attention map by a specific ratio. When the processor 110 increases the size of the spatial attention map by a specific ratio, data of a newly created array element may be filled with 0 or a predetermined number. The term "specific ratio" may be used interchangeably with the word "stride" to have the same meaning. For example, if a 2D array having a size of 2x2 is upsampled by a stride of 32, the 2D array has a size of 64x64. If the data of the newly created array element is filled with zeros, the upsampled 64x64 array may include a total of four array elements having non-zero data. The above-described example is merely an example and does not limit the present disclosure. In the present disclosure, the processor 110 performs upsampling as in the above-described example to make the two arrays the same size even when the spatial attention map and the attention guide label have different sizes, and then compares them with each other and then based on the comparison result. A bone age analysis model can be trained.

The supervised learning of the bone age analysis model performed by the processor 110 in an embodiment of the present disclosure is a result calculated by substituting a spatial attention map and an attention guide label into a binary cross-entropy loss function. can be performed based on The spatial attention map may be used as a prediction value in the binary cross entropy loss function. The attention guide label may be used as a true value in the binary cross entropy loss function. The binary cross entropy loss function may be calculated and summed for each pixel of the spatial attention map. The binary cross entropy loss function can be learned so that the result value is close to zero. The equation for the binary cross entropy loss function is as Equation (4).

[Equation 4]

는 어텐션 가이드 라벨에 포함된

번째 픽셀의 라벨을 나타낸다. 상기 라벨은 0 또는 1로 표현될 수 있으며 이진 분류(Binary Classification)을 위해 적어도 하나 이상의 픽셀의 중요도가 사전 결정된 임계값보다 클 경우 1 또는 사전 결정된 임계값보다 작을 경우 0으로 결정될 수 있다. 상기 이진 분류를 위하 사전 결정된 임계값은 사용자에 의해 임의로 설정될 수 있다. 상기 이진 분류를 위한 사전 결정된 임계값은 하나 이상의 픽셀의 중요도 값들의 전체 분포 함수를 정규화한 후 정규 분포 함수에서 상위 M%에 해당하는 값으로 설정될 수도 있다. 상기 M은 25, 50 등과 같이 사용자에 의해 설정될 수 있다. 2차원 배열로 이루어진 공간 어텐션 맵 또는 어텐션 가이드 라벨에 포함된 적어도 하나 이상의 픽셀은 임의의 기준에 따라 순서가 부여될 수 있다. 상기 수학식 4의

는 프로세서(110)가 골 연령 분석 모델에 포함된 어텐션 모듈을 이용하여 학습 영상에 대해서 생성한 공간 어텐션 맵에 포함된

번째 픽셀의 예측 중요도를 나타낸다. 상기 예측 중요도는 정규화되어 0과 1 사이의 값을 가지도록 설정될 수 있다. 다른 실시예로 프로세서(110)는 손실 함수로써 이진 크로스 엔트로피 함수가 아닌 크로스 엔트로피 함수를 사용할 수 있다. 본 개시에 있어서 프로세서(110)는 상술한 손실함수의 손실 값이 최소화되도록 어텐션 모듈에 포함된 채널 어텐션 신경망 모델 및 공간 어텐션 신경망 모델 내부 연결 가중치 및 편향값을 갱신함으로써 어텐션 모듈을 학습시킬 수 있다.in Equation 4

is included in the attention guide label.

Indicates the label of the second pixel. The label may be expressed as 0 or 1, and may be determined as 1 when the importance of at least one or more pixels for binary classification is greater than a predetermined threshold or 0 when less than the predetermined threshold. The predetermined threshold for the binary classification may be arbitrarily set by a user. The predetermined threshold value for the binary classification may be set to a value corresponding to the upper M% in the normal distribution function after normalizing the entire distribution function of importance values of one or more pixels. The M may be set by the user, such as 25, 50, or the like. At least one or more pixels included in the spatial attention map or the attention guide label having a two-dimensional arrangement may be ordered according to an arbitrary criterion. of Equation 4 above

is included in the spatial attention map generated for the learning image by the processor 110 using the attention module included in the bone age analysis model.

Indicates the prediction importance of the th pixel. The prediction importance may be normalized and set to have a value between 0 and 1. In another embodiment, the processor 110 may use a cross entropy function instead of a binary cross entropy function as the loss function. In the present disclosure, the processor 110 may learn the attention module by updating the internal connection weights and bias values of the channel attention neural network model and the spatial attention neural network model included in the attention module so that the loss value of the above-described loss function is minimized.

In an embodiment of the present disclosure, the supervised learning of the bone age analysis model is a predetermined attention module based on the result value of the loss function calculated in each attention module, when the bone age analysis model includes at least two or more attention modules. It may be performed based on the summed result after multiplying the weights. As an example for explanation, the equation is the same as Equation 5.

[Equation 5]

은 전체 손실값을 나타낸다.

는 골 연령 분석 모델이 예측한 골 연령과 실제 나이의 차이를 고려한 손실값을 나타낸다.

는

번째 어텐션 모듈이 생성한 공간 어텐션 맵과 어텐션 가이드 라벨의 비교 결과에 기초한 손실값을 나타낸다.

는

번째 어텐션 모듈의 손실값에 대한 어텐션 모듈에 따른 가중치를 나타낸다. 상기한 바와 같이 본 개시에 있어서 프로세서(110)는 골 연령 분석 모델을 학습시킬 때, 최종 출력에 따른 예측 연령에 기초한 회귀 손실값 뿐만 아니라 적어도 하나 이상의 어텐션 모듈이 생성하는 공간 어텐션 맵에 기초한 손실값을 고려하여 학습을 진행하므로 골 연령 분석 모델로 하여금 골 영상의 각 부분 중 주요 영역을 중심으로 분석을 수행하게 할 수 있다. 그 결과 전체 골 연령 판독 성능이 향상되는 효과를 가진다. 또한 본 개시에 있어서 프로세서(110)는 골 연령 분석 모델이 둘 이상의 어텐션 모듈을 포함하는 경우, 어텐션 모듈에 따른 가중치를 두어 전체 손실값을 결정할 수 있다. 즉, 프로세서(110)는 지도학습을 위한 전체 손실값을 결정하는데 있어서 어텐션 모듈별로 영향을 조절할 수 있다. 본 개시의 골 연령 분석 모델에 있어서 어텐션 모듈이 복수 개 존재하는 경우, 전체 모델의 출력단에 가까이 위치하는 어텐션 모듈일수록 연산 과정에서 전체 모델의 입력단에 가까이 위치하는 어텐션 모듈보다 입력된 학습 영상의 세부 위치 정보를 잃어버렸을 가능성이 높다. 위와 같은 경우 주요 영역을 중점적으로 분석하기 위한 공간 어텐션 맵의 생성에 있어서 입력된 학습 영상의 세부 위치 정보를 보다 정확하게 알고 있는 전체 모델의 입력단 쪽에 위치하는 어텐션 모듈의 손실을 보다 높게 고려하도록 가중치를 고려할 수 있다. 이렇게 프로세서(110)는 어텐션 모듈에 따른 가중치를 두어 전체 손실값을 결정하는데 있어서 어텐션 모듈별로 그 영향을 조절하므로 효과적인 학습 결과를 얻을 수 있다.of Equation 5

is the total loss value.

represents the loss value considering the difference between the bone age predicted by the bone age analysis model and the actual age.

Is

A loss value based on a comparison result between the spatial attention map generated by the th attention module and the attention guide label is shown.

Is

It represents a weight according to the attention module for the loss value of the th attention module. As described above, in the present disclosure, when the processor 110 trains the bone age analysis model, the loss value based on the spatial attention map generated by at least one attention module as well as the regression loss value based on the predicted age according to the final output. Since learning is carried out in consideration of As a result, the overall bone age reading performance is improved. In addition, in the present disclosure, when the bone age analysis model includes two or more attention modules, the processor 110 may determine the total loss value by assigning weights according to the attention modules. That is, the processor 110 may adjust the influence for each attention module in determining the overall loss value for supervised learning. When there are a plurality of attention modules in the bone age analysis model of the present disclosure, the closer the attention module is located to the output end of the entire model, the more detailed the position of the input learning image is compared to the attention module located closer to the input end of the entire model in the calculation process. It is highly likely that information has been lost. In the above case, in the generation of the spatial attention map for intensive analysis of the main area, the weight is considered to be higher to consider the loss of the attention module located at the input end of the entire model that knows the detailed location information of the input training image more accurately. can In this way, the processor 110 adjusts the influence for each attention module in determining the overall loss value by assigning a weight according to the attention module, so that an effective learning result can be obtained.

In an embodiment of the present disclosure, the processor 110 of the computing device displays a user interface screen including a heat map generated based on the spatial attention map for the analyzed image generated by the attention module included in the bone age analysis model. can provide After the learning of the bone age analysis model is completed, the computing device 100 may receive the analysis image. The analysis image may be a new image that does not exist in the learning image. In this case, the above-described attention guide label may not exist for the analyzed image. The processor 110 may generate a heat map based on the spatial attention map for the analyzed image. The processor 110 may provide a user interface screen including the heat map to the user. The heat map may be a visual representation of a spatial attention map generated by the attention module while the bone age analysis model calculates to read the bone age from the inputted analysis image. The visual representation of the heat map may be a color-based representation according to the size of the spatial attention map internal parameter value. For example, the larger the parameter value, the closer to red, and the smaller the parameter value, the closer to blue. The parameter may be a value calculated based on the importance of the corresponding pixel. The example related to the above-described heat map representation is only an example and does not limit the present disclosure. The heat map may be included in a portion of the user interface screen and provided to the user.

In an embodiment of the present disclosure, the computing device may further include an output unit (not shown). The output unit (not shown) includes a liquid crystal display (LCD), a thin film transistor-liquid crystal display (TFT LCD), an organic light-emitting diode (OLED), a flexible display ( flexible display) and at least one of a three-dimensional display (3D display). Some of these display modules may be of a transparent type or a light transmission type so that the outside can be seen through them. This may be referred to as a transparent display module, and a representative example of the transparent display module is a transparent OLED (TOLED). The computing device of the present disclosure provides a user interface screen including a heat map generated based on the spatial attention map for the analyzed image generated by the processor 110 to an output unit (not shown) to display it so that the user can see it. can do. Also, the computing device of the present disclosure may provide a user interface screen including a heat map generated based on the spatial attention map for the analyzed image generated by the processor 110 to an external device. The computing device of the present disclosure may transmit the user interface screen to an external device through the network unit 150 so that the external device displays the screen to the user.

The bone age reading method of the present disclosure provides the user with a user interface screen including a heat map generated based on the spatial attention map for the analyzed image, thereby allowing the user to improve the reliability of the analysis model's result inference and to improve the transparency of the model. has the effect of enhancing Hereinafter, it will be described by way of example with reference to FIG. 6 .

6 is an exemplary diagram illustrating a user interface screen including a heat map generated based on a spatial attention map for an analysis image according to an embodiment of the present disclosure.

Reference numeral 630 of FIG. 6 includes a heat map that visually represents the spatial attention map generated by the bone age analysis model including the supervised attention module. Referring to the area of reference numeral 630, when the bone age analysis model includes, for example, three attention modules, the bone age analysis model calculates a total of three spatial attention maps for the first bone image 631 in the calculation process. can create Heat maps visually representing the plurality of spatial attention maps may be expressed as color-based heat maps according to importance. For example, the heat map visually representing the spatial attention map may be displayed by setting a color, brightness, or density differently according to a difference in importance values. As an example, in pixels included in the heat map, a pixel having a higher importance may be displayed in red color, and a corresponding pixel may be displayed in blue color as the importance level is lower. As another example, a pixel included in the heat map may be displayed brightly as its importance is high and darkly as its importance is low. The description of the above-described expression method of the heat map is only an example and does not limit the present disclosure. Referring to FIG. 6 , as the first heat map 633 , the second heat map 635 , and the third heat map 637 for the plurality of spatial attention maps proceed in the order, the closer to the output end of the bone age analysis model, the degree of abstraction It can be seen that the higher the value, the lower the resolution. Looking at the first heat map 633 , the second heat map 635 , or the third heat map 637 , the bone age analysis model assigns high importance to the knuckle or joint part, which is a major area for bone age reading in the carpal image. It can be seen that the analysis is The processor 110 may provide a user interface screen including at least one of the first heat map 633 , the second heat map 635 , and the third heat map 637 to the user. At this time, the user may check the spatial attention map and check which part of the bone age analysis model viewed the first bone image 631 to evaluate the bone age.

Reference numeral 610 of FIG. 6 includes a heat map that visually represents the spatial attention map generated by the bone age analysis model including the unsupervised attention module. For example, the bone age analysis model unsupervised for the second marrow image 611 may generate three spatial attention maps, and the heat maps for the plurality of attention maps are the fourth heat map 613, A fifth heat map 615 and a sixth heat map 617 may be displayed. This shows a contrasting result when compared with the first heat map 633 , the second heat map 635 , and the third heat map 637 derived by the supervised bone age analysis model of the present disclosure. Specifically, most of the pixels of reference number 613 are expressed in green, which indicates that the first unsupervised learning attention module judges almost all areas of the bone image as main areas, and gives high importance to most of the pixels. That is, even if the bone age analysis model includes the attention module, if it is not trained to analyze the main areas by supervised learning using the attention guide label as in the present disclosure, the bone age analysis model is not analyzed in evaluating the age of the bones. Computing resources are used in areas that are not needed. Therefore, the bone age analysis model including the supervised attention module according to an embodiment of the present disclosure has the advantage of effectively learning the model and improving performance.

7 is a flowchart for a computing device to read a bone age for an analyzed image and provide a user interface screen according to an embodiment of the present disclosure. The network unit 150 of the computing device 100 of the present disclosure may receive 710 an analysis image that is a target of bone age reading. The analysis image may be an image different from the previously learned learning image. The processor 110 of the computing device 100 may input the analyzed image into a bone age analysis model including one or more neural networks to read 730 bone age. The bone age analysis model may include at least one attention module for focusing on a main area. The attention module may include a channel attention neural network model for generating a channel attention map and a spatial attention neural network model for generating a spatial attention map. Thereafter, the processor 110 of the computing device 100 provides a user interface screen including a heat map generated based on the spatial attention map for the analyzed image generated by the attention module included in the bone age analysis model. (750) can. By checking the spatial attention map for the analyzed image, the user has the effect of easily knowing which part was viewed and the conclusion of the bone age analysis model was drawn.

8 is a simplified, general schematic diagram of an exemplary computing environment in which embodiments of the present disclosure may be implemented.

Although the present disclosure has been described above generally as being capable of being implemented by a computing device, those skilled in the art will appreciate that the present disclosure is a combination of hardware and software and/or in combination with computer-executable instructions and/or other program modules that may be executed on one or more computers. It will be appreciated that it can be implemented as a combination.

Generally, program modules include routines, programs, components, data structures, etc. that perform particular tasks or implement particular abstract data types. In addition, those skilled in the art will appreciate that the methods of the present disclosure can be applied to single-processor or multiprocessor computer systems, minicomputers, mainframe computers as well as personal computers, handheld computing devices, microprocessor-based or programmable consumer electronics, and the like. It will be appreciated that each of these may be implemented in other computer system configurations, including those capable of operating in connection with one or more associated devices.

The described embodiments of the present disclosure may also be practiced in distributed computing environments where certain tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.

Computers typically include a variety of computer-readable media. Any medium accessible by a computer can be a computer readable medium, and such computer readable media includes volatile and nonvolatile media, transitory and non-transitory media, removable and non-transitory media. including removable media. By way of example, and not limitation, computer-readable media may include computer-readable storage media and computer-readable transmission media. Computer-readable storage media includes volatile and non-volatile media, temporary and non-transitory media, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. includes media. A computer-readable storage medium may be RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital video disk (DVD) or other optical disk storage device, magnetic cassette, magnetic tape, magnetic disk storage device, or other magnetic storage device. device, or any other medium that can be accessed by a computer and used to store the desired information.

Computer readable transmission media typically embodies computer readable instructions, data structures, program modules or other data, etc. in a modulated data signal such as a carrier wave or other transport mechanism, and Includes any information delivery medium. The term modulated data signal means a signal in which one or more of the characteristics of the signal is set or changed so as to encode information in the signal. By way of example, and not limitation, computer-readable transmission media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, RF, infrared, and other wireless media. Combinations of any of the above are also intended to be included within the scope of computer-readable transmission media.

An exemplary environment 1100 implementing various aspects of the disclosure is shown including a computer 1102 , the computer 1102 including a processing unit 1104 , a system memory 1106 , and a system bus 1108 . do. A system bus 1108 couples system components, including but not limited to system memory 1106 , to the processing device 1104 . The processing device 1104 may be any of a variety of commercially available processors. Dual processor and other multiprocessor architectures may also be used as processing unit 1104 .

The system bus 1108 may be any of several types of bus structures that may be further interconnected to a memory bus, a peripheral bus, and a local bus using any of a variety of commercial bus architectures. System memory 1106 includes read only memory (ROM) 1110 and random access memory (RAM) 1112 . A basic input/output system (BIOS) is stored in non-volatile memory 1110, such as ROM, EPROM, EEPROM, etc., which BIOS is the basic input/output system (BIOS) that helps transfer information between components within computer 1102, such as during startup. contains routines. RAM 1112 may also include high-speed RAM, such as static RAM, for caching data.

The computer 1102 may also be configured with an internal hard disk drive (HDD) 1114 (eg, EIDE, SATA) - this internal hard disk drive 1114 may also be configured for external use within a suitable chassis (not shown). Yes-, magnetic floppy disk drive (FDD) 1116 (eg, for reading from or writing to removable diskette 1118), and optical disk drive 1120 (eg, CD-ROM) for reading from, or writing to, disk 1122, or other high capacity optical media such as DVD. The hard disk drive 1114 , the magnetic disk drive 1116 , and the optical disk drive 1120 are connected to the system bus 1108 by the hard disk drive interface 1124 , the magnetic disk drive interface 1126 , and the optical drive interface 1128 , respectively. ) can be connected to The interface 1124 for implementing an external drive includes at least one or both of Universal Serial Bus (USB) and IEEE 1394 interface technologies.

These drives and their associated computer readable media provide non-volatile storage of data, data structures, computer executable instructions, and the like. For computer 1102, drives and media correspond to storing any data in a suitable digital format. Although the description of computer readable media above refers to HDDs, removable magnetic disks, and removable optical media such as CDs or DVDs, those skilled in the art will use zip drives, magnetic cassettes, flash memory cards, cartridges, etc. It will be appreciated that other tangible computer-readable media such as etc. may also be used in the exemplary operating environment and any such media may include computer-executable instructions for performing the methods of the present disclosure.

A number of program modules may be stored in the drive and RAM 1112 , including an operating system 1130 , one or more application programs 1132 , other program modules 1134 , and program data 1136 . All or portions of the operating system, applications, modules, and/or data may also be cached in RAM 1112 . It will be appreciated that the present disclosure may be implemented in various commercially available operating systems or combinations of operating systems.

A user may enter commands and information into the computer 1102 via one or more wired/wireless input devices, for example, a pointing device such as a keyboard 1138 and a mouse 1140 . Other input devices (not shown) may include a microphone, IR remote control, joystick, game pad, stylus pen, touch screen, and the like. Although these and other input devices are often connected to the processing unit 1104 through an input device interface 1142 that is connected to the system bus 1108, parallel ports, IEEE 1394 serial ports, game ports, USB ports, IR interfaces, It may be connected by other interfaces, etc.

A monitor 1144 or other type of display device is also coupled to the system bus 1108 via an interface, such as a video adapter 1146 . In addition to the monitor 1144, the computer typically includes other peripheral output devices (not shown), such as speakers, printers, and the like.

Computer 1102 may operate in a networked environment using logical connections to one or more remote computers, such as remote computer(s) 1148 via wired and/or wireless communications. Remote computer(s) 1148 may be workstations, computing device computers, routers, personal computers, portable computers, microprocessor-based entertainment devices, peer devices, or other common network nodes, and are typically connected to computer 1102 . Although it includes many or all of the components described for it, only memory storage device 1150 is shown for simplicity. The logical connections shown include wired/wireless connections to a local area network (LAN) 1152 and/or a larger network, eg, a wide area network (WAN) 1154 . Such LAN and WAN networking environments are common in offices and companies, and facilitate enterprise-wide computer networks, such as intranets, all of which can be connected to a worldwide computer network, for example, the Internet.

When used in a LAN networking environment, the computer 1102 is connected to the local network 1152 through a wired and/or wireless communication network interface or adapter 1156 . Adapter 1156 may facilitate wired or wireless communication to LAN 1152 , which also includes a wireless access point installed therein for communicating with wireless adapter 1156 . When used in a WAN networking environment, the computer 1102 may include a modem 1158, be connected to a communication computing device on the WAN 1154, or establish communications over the WAN 1154, such as over the Internet. have other means. A modem 1158 , which may be internal or external and a wired or wireless device, is coupled to the system bus 1108 via a serial port interface 1142 . In a networked environment, program modules described for computer 1102 or portions thereof may be stored in remote memory/storage device 1150 . It will be appreciated that the network connections shown are exemplary and other means of establishing a communication link between the computers may be used.

Computer 1102 may be associated with any wireless device or object that is deployed and operates in wireless communication, for example, printers, scanners, desktop and/or portable computers, portable data assistants (PDAs), communication satellites, wireless detectable tags. It operates to communicate with any device or place, and phone. This includes at least Wi-Fi and Bluetooth wireless technologies. Accordingly, the communication may be a predefined structure as in a conventional network or may simply be an ad hoc communication between at least two devices.

Wi-Fi (Wireless Fidelity) makes it possible to connect to the Internet, etc. without a wire. Wi-Fi is a wireless technology such as cell phones that allows these devices, eg, computers, to transmit and receive data indoors and outdoors, ie anywhere within range of a base station. Wi-Fi networks use a radio technology called IEEE 802.11 (a, b, g, etc) to provide secure, reliable, and high-speed wireless connections. Wi-Fi can be used to connect computers to each other, to the Internet, and to wired networks (using IEEE 802.3 or Ethernet). Wi-Fi networks may operate in unlicensed 2.4 and 5 GHz radio bands, for example, at 11 Mbps (802.11a) or 54 Mbps (802.11b) data rates, or in products that include both bands (dual band). .

One of ordinary skill in the art of this disclosure will understand that information and signals may be represented using any of a variety of different technologies and techniques. For example, data, instructions, instructions, information, signals, bits, symbols, and chips that may be referenced in the above description are voltages, currents, electromagnetic waves, magnetic fields or particles, optical field particles or particles, or any combination thereof.

A person of ordinary skill in the art of the present disclosure will recognize that the various illustrative logical blocks, modules, processors, means, circuits and algorithm steps described in connection with the embodiments disclosed herein include electronic hardware, (convenience For this purpose, it will be understood that it may be implemented by various forms of program or design code (referred to herein as software) or a combination of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. A person skilled in the art of the present disclosure may implement the described functionality in various ways for each specific application, but such implementation decisions should not be interpreted as a departure from the scope of the present disclosure.

The various embodiments presented herein may be implemented as methods, apparatus, or articles of manufacture using standard programming and/or engineering techniques. The term article of manufacture includes a computer program, carrier, or media accessible from any computer-readable storage device. For example, computer-readable storage media include magnetic storage devices (eg, hard disks, floppy disks, magnetic strips, etc.), optical disks (eg, CDs, DVDs, etc.), smart cards, and flash drives. memory devices (eg, EEPROMs, cards, sticks, key drives, etc.). Also, various storage media presented herein include one or more devices and/or other machine-readable media for storing information.

It is understood that the specific order or hierarchy of steps in the presented processes is an example of exemplary approaches. Based on design priorities, it is to be understood that the specific order or hierarchy of steps in the processes may be rearranged within the scope of the present disclosure. The appended method claims present elements of the various steps in a sample order, but are not meant to be limited to the specific order or hierarchy presented.

The description of the presented embodiments is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the scope of the present disclosure. Thus, the present disclosure is not intended to be limited to the embodiments presented herein, but is to be construed in the widest scope consistent with the principles and novel features presented herein.

Claims (12)

A method of reading bone age using a neural network performed by a computing device, comprising:
Receiving an analysis image that is a target of bone age reading; and
reading the bone age by inputting the analyzed image into a bone age analysis model including one or more neural networks;
including, and
The bone age analysis model is,
At least one attention module (Attention Module) for centrally analyzing the main area of the analyzed image, characterized in that supervised learning is performed based on the attention guide label (Attention Guide Label),
Bone age reading method using neural network.
The method of claim 1,
The attention guide label is
It is generated based on the detection result obtained as a result of inputting the learning image into a main region detection model including at least one or more neural networks,
The detection result includes location information on at least one main area included in the learning image,
Bone age reading method using neural network.
3. The method of claim 2,
When the main area is in the form of a bounding box,
The detection result is
Including the coordinates of the center point of at least one main area included in the learning image, the width of the main area and the height of the main area,
Bone age reading method using neural network.
The method of claim 1,
The attention guide label is
In at least one pixel included in the training image,
Including the importance of the pixel obtained as a result of substituting the distance between the coordinates of the pixel and the coordinates of the center point of one or more major regions included in the training image into an equation based on a Gaussian distribution,
Bone age reading method using neural network.
The method of claim 1,
The bone age analysis model is,
at least one learning image; and
an attention guide label corresponding to each of the learning images and including at least one main area;
supervised learning based on
The supervised learning is
It is performed based on a comparison result of a spatial attention map generated for the learning image using the bone age analysis model and an attention guide label corresponding to the learning image,
Bone age reading method using neural network.
6. The method of claim 5,
The supervised learning is
Performed based on a result calculated by substituting the spatial attention map and the attention guide label into a binary cross-entropy loss function,
Bone age reading method using neural network.
The method of claim 1,
Supervised learning for the bone age analysis model,
When the bone age analysis model includes at least two or more attention modules,
It is performed based on the summed result after multiplying the result value of the loss function calculated in each attention module by a weight according to the predetermined attention module,
Bone age reading method using neural network.
The method of claim 1,
The attention module is
a channel attention neural network model for generating a channel attention map with respect to the feature map input to the attention module; and
a spatial attention neural network model for generating a spatial attention map for the modified feature map;
including,
The modified feature map is
A feature map in which the channel attention map is multiplied by element by the feature map input to the attention module,
Bone age reading method using neural network.
A method of reading bone age using a neural network performed by a computing device, comprising:
Receiving an analysis image that is a target of bone age reading;
reading the bone age by inputting the analyzed image into a bone age analysis model including one or more neural networks; and
providing a user interface screen including a heat map generated based on a spatial attention map for the analyzed image;
including, and
The bone age analysis model is,
At least one attention module (Attention Module) for centrally analyzing the main area of the analyzed image, characterized in that supervised learning is performed based on the attention guide label (Attention Guide Label),
Bone age reading method using neural network.
A method of reading bone age using a neural network performed by a computing device, comprising:
Receiving an analysis image that is a target of bone age reading;
reading the bone age by inputting the analyzed image into a bone age analysis model including one or more neural networks; and
providing a user interface screen including a heat map generated based on a spatial attention map for the analyzed image;
including, and
The attention module supervised based on the attention guide label included in the bone age analysis model (Attention Module),
a channel attention neural network model for generating a channel attention map with respect to the feature map input to the attention module; and
a spatial attention neural network model for generating a spatial attention map for the modified feature map;
includes,
The modified feature map is
A feature map in which the channel attention map is multiplied by element by the feature map input to the attention module,
Bone age reading method using neural network.
A computer program stored in a non-transitory computer-readable storage medium, wherein the computer program, when executed on one or more processors, performs the following operations for reading bone age using a neural network, the operations comprising:
reading the age of the bones by inputting the analyzed image, which is the target of bone age reading, into a bone age analysis model including one or more neural networks;
including, and
The bone age analysis model is,
At least one attention module (Attention Module) for centrally analyzing the main area of the analyzed image, characterized in that supervised learning is performed based on the attention guide label (Attention Guide Label),
A computer program stored in a non-transitory computer-readable storage medium.
A device for reading bone age, comprising:
one or more processors; and
a memory storing a bone aging model comprising one or more neural networks;
a network unit for receiving an analysis image that is a target of bone age reading;
including, and
The one or more processors,
By inputting the analysis image into a bone age analysis model including one or more neural networks, the bone age is read,
The bone age analysis model is,
At least one attention module (Attention Module) for centrally analyzing the main area of the analyzed image, characterized in that supervised learning is performed based on the attention guide label (Attention Guide Label),
A device for reading bone age.

Images