JP7318058B2 - Image processing device - Google Patents

Description

An embodiment of the present invention relates to an image processing apparatus .

Techniques for automatically identifying texture patterns in images acquired by a medical image diagnostic apparatus such as an X-ray CT (Computed Tomography) apparatus are known. This type of technique is useful for diagnosing diseases such as diffuse lung disease, which exhibit characteristic texture patterns on CT images (see, for example, Non-Patent Document 1). Therefore, this type of technology is expected to be utilized for diagnosis of predetermined diseases.

By the way, a doctor performs radiogram interpretation by extracting a characteristic texture pattern from a patient's CT image. With recent improvements in the resolution of X-ray CT apparatuses, etc., it has become possible to interpret lesions in detail, but visual interpretation is a heavy burden on doctors. In addition, since experience is required to extract a characteristic texture pattern, the diagnosis results may vary from doctor to doctor.

In order to solve such problems, computer aided diagnosis has attracted attention in recent years. For example, a method is known in which feature values are extracted from a region of interest belonging to the lung field of a CT image, and texture patterns are automatically identified by machine learning based on the extracted feature values. The design of the feature extraction method is important to improve the identification accuracy. However, since the feature amount is directly extracted from the CT image, it is difficult to extract the feature amount that is effective for identification, and it is sometimes difficult to improve the accuracy of the identification.

U.S. Pat. No. 7,236,619

D. M. Hansell, "Fleischner Society: glossary of terms for thoracic imaging", Radiology, Vol. 246, No. 3, p. 697-722, 2008 V. Lepetit, "Keypoint Recognition using Randomized Trees," IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 28, No. 9, p. 1465-1479, 2006 A. F. Frangi, "Multiscale Vessel Enhancement Filtering", Medical Image Computing and Computer-Assisted Intervention (MICCAI), LNCS, vol. 1496, p. 130-137, 1998

Therefore, the object is to improve the identification accuracy of texture patterns in medical images.

According to an embodiment, an image processing apparatus includes an image acquisition section, a likelihood acquisition section, and a feature amount calculation section. The image acquisition unit acquires medical image data. A likelihood acquisition unit obtains a plurality of likelihood values representing the likelihood that an element included in a medical image based on the medical image data is classified into at least one of classification items corresponding to substances and structures in the body. Acquire for each pixel of the medical image. The feature amount calculator calculates a plurality of feature amounts using the plurality of likelihood values.

FIG. 1 is a diagram showing a medical information system including an image processing apparatus according to this embodiment. FIG. 2 is a diagram showing a functional configuration of the image processing apparatus shown in FIG. 1; FIG. 3 is a diagram for explaining the operation of the processing circuit shown in FIG. 2; FIG. 4 is a diagram showing a process of obtaining likelihood values by the likelihood obtaining function shown in FIG. FIG. 5 is a diagram illustrating another example of processing for obtaining likelihood values by the likelihood obtaining function illustrated in FIG. 2 . FIG. 6 is a diagram showing distances between pixel positions. FIG. 7 is a diagram showing an example of processing for acquiring likelihood values through decision tree learning by the likelihood acquisition function shown in FIG. FIG. 8 is a diagram showing an example of processing for acquiring a likelihood value regarding a structure by a structure-emphasizing filter using the likelihood acquisition function shown in FIG. 2 . FIG. 9 is a diagram showing the process of creating a feature vector by the feature amount calculation function shown in FIG. FIG. 10 is a diagram showing processing for learning weighting factors using a neural network. FIG. 11 is a diagram showing an example in which the functional configuration of the image processing apparatus shown in FIG. 1 has identification capability. 12 is a diagram for explaining another example of the operation of the processing circuit shown in FIG. 2; FIG. FIG. 13 is a diagram showing the functional configuration of the medical image diagnostic apparatus shown in FIG. 1;

Embodiments will be described below with reference to the drawings.

FIG. 1 is a diagram showing an example of a medical information system 1 including an image processing apparatus 10 according to this embodiment. The medical information system 1 shown in FIG. 1 comprises an image processing device 10 , a medical image diagnostic device 20 and an image storage device 30 . The image processing apparatus 10, the medical image diagnostic apparatus 20, and the image storage apparatus 30 are directly or indirectly connected to communicate with each other via an in-hospital LAN (Local Area Network) installed in the hospital, for example. there is For example, when the image archiving device 30 constitutes a PACS (Picture Archiving and Communication System), the image processing device 10, the medical image diagnostic device 20, and the image archiving device 30 conform to the DICOM (Digital Imaging and Communications in Medicine) standard. For example, they exchange medical image data with each other.

The medical image diagnostic apparatus 20 is an apparatus that generates medical image data by photographing a subject. The medical image diagnostic device 20 includes, for example, an X-ray diagnostic device, an X-ray CT (Computed Tomography) device, an MRI (Magnetic Resonance Imaging) device, an ultrasonic diagnostic device, a SPECT (Single Photon Emission Computed Tomography) device, a PET (Positron Emission computed tomography) device, SPECT-CT device in which SPECT device and X-ray CT device are integrated, PET-CT device in which PET device and X-ray CT device are integrated, PET device and MRI device in one a PET-MRI device, or a group of these devices.

The image storage device 30 is a database that stores medical image data. The image storage device 30 stores, for example, medical image data generated by the image processing device 10 and the medical image diagnostic device 20 in a storage circuit provided therein.

The image processing apparatus 10 performs image processing on medical image data generated by the medical image diagnostic apparatus 20 and medical image data read from the image storage apparatus 30 . FIG. 2 is a diagram showing an example of the functional configuration of the image processing apparatus 10 according to this embodiment. The image processing apparatus 10 shown in FIG. 2 has a processing circuit 11 , an input interface 12 , an output interface 13 , a communication interface 14 and a memory circuit 15 .

The processing circuit 11 is a processor that functions as the core of the image processing apparatus 10 . By executing the program stored in the storage circuit 15, the processing circuit 11 realizes a function corresponding to the executed program. Note that the processing circuit 11 may have a storage area for storing at least part of the data stored in the storage circuit 15 .

The input interface 12 receives various operations input by the operator to the image processing apparatus 10 . The input interface 12 is implemented by, for example, a mouse, a keyboard, and a touch panel through which an instruction is input by touching an operation surface. The input interface 12 is connected to the processing circuit 11 , converts an operation instruction input by an operator into an electrical signal, and outputs the electrical signal to the processing circuit 11 . It should be noted that the input interface 12 in this specification is not limited to having physical operation parts such as a mouse and a keyboard. For example, an electrical signal processing circuit that receives an electrical signal corresponding to an operation instruction input from an external input device provided separately from the image processing apparatus 10 and outputs this electrical signal to the processing circuit 11 is also an input interface. Included in 12 examples.

The output interface 13 is connected to the processing circuit 11 and outputs signals supplied from the processing circuit 11 . The output interface 13 is, for example, a display circuit, and is realized by, for example, a CRT display, liquid crystal display, organic EL display, LED display, plasma display, or the like. The display circuit displays, for example, a medical image based on medical image data. The output interface 13 also includes a processing circuit that converts data representing an object to be displayed into a video signal and outputs the video signal to the outside.

The communication interface 14 connects with, for example, an intra-hospital network. The communication interface 14 receives medical image data from the medical image diagnostic apparatus 20 and the image storage apparatus 30 via, for example, an intra-hospital network.

The memory circuit 15 includes a processor-readable recording medium such as a magnetic or optical recording medium, or a semiconductor memory. Also, the storage circuit 15 may be a drive device or the like that reads and writes various information from/to a portable storage medium such as a CD-ROM drive, a DVD drive, and a flash memory. Note that the storage circuit 15 does not necessarily have to be realized by a single storage device. For example, the memory circuit 15 may be realized by a plurality of memory devices.

The storage circuit 15 stores the received data under the control of the processing circuit 11 . For example, the storage circuit 15 stores medical image data output from the medical image diagnostic apparatus 20 and the image storage apparatus 30 .

Further, the storage circuit 15 reads stored data according to control from the processing circuit 11 . For example, the memory circuit 15 reads the stored medical image data under the control of the processing circuit 11 . Also, for example, the storage circuit 15 stores a program according to the present embodiment. The storage circuit 15 reads the stored program under the control of the processing circuit 11 . The memory circuit 15 also stores, for example, data on frequency distribution of substances and structures in the body, data on decision tree learning, data on structure enhancement filters, data on weighting factors, and the like. The storage circuit 15 reads various stored data according to control from the processing circuit 11 .

By executing the program according to the present embodiment, the processing circuit 11 according to the present embodiment calculates the feature quantity using the likelihood that the elements included in the medical image are classified into the classification item corresponding to the predetermined configuration. do. Specifically, the processing circuit 11 has an image acquisition function 110 , a likelihood acquisition function 111 , and a feature quantity calculation function 112 by executing programs stored in the storage circuit 15 .

The image acquisition function 110 is a function for acquiring desired medical image data. For example, when the image acquisition function 110 is executed, the processing circuit 11 reads medical image data stored in the storage circuit 15 . Note that the image acquisition function 110 may acquire desired medical image data from the medical image diagnostic apparatus 20 and the image storage apparatus 30 .
The likelihood acquisition function 111 is a function for acquiring a likelihood representing the likelihood that an element included in a medical image will be classified into a classification item corresponding to a predetermined configuration. For example, when the likelihood acquisition function 111 is executed, the processing circuitry 11 acquires N likelihood values for each pixel of the medical image to generate a multi-channel image. FIG. 3 is a diagram schematically explaining the operation of the processing circuit 11 according to this embodiment. According to FIG. 3, the processing circuit 11 generates likelihood images of likelihood 1 to likelihood N by obtaining a likelihood value for each pixel of the medical image.

The feature amount calculation function 112 is a function for calculating feature amounts based on the likelihood values acquired by the likelihood acquisition function 111 . For example, when the feature amount calculation function 112 is executed, the processing circuit 11 calculates a plurality of feature amounts using the obtained plurality of likelihood values, and generates a feature vector from the calculated plurality of feature amounts. According to FIG. 3, the processing circuit 11 generates a feature vector based on likelihood images of likelihood 1 to likelihood N. FIG.

Next, the processing of the likelihood acquisition function 111 shown in FIG. 2 will be specifically described. In the following description, the medical image diagnostic apparatus 20 is an X-ray CT apparatus that generates CT image data, and the image processing apparatus 10 receives the CT image data generated by the medical image diagnostic apparatus 20 as medical image data. A case will be described as an example.

First, the medical image diagnostic apparatus 20, which is an X-ray CT apparatus, images an imaging region of a subject with X-rays. Specifically, the medical image diagnostic apparatus 20 generates X-rays from the X-ray tube while rotating a rotating frame to which the X-ray tube and the X-ray detector are attached. The X-ray detector detects X-rays generated from the X-ray tube and transmitted through the subject. The medical image diagnostic apparatus 20 acquires raw data corresponding to X-rays detected by an X-ray detector using a data acquisition system (DAS), and produces CT image data based on the acquired raw data. Reconstructed by an image reconstruction device.

It is assumed that the CT image data according to the present embodiment is data representing a slice image showing a two-dimensional spatial distribution of CT values. A slice image is composed of a plurality of pixels arranged two-dimensionally. Each pixel is assigned a CT value. A CT image based on CT image data may be an image obtained by imaging the entire target organ, or may be an image limited to a local region of interest. The CT image data may be data representing a volume image showing a three-dimensional spatial distribution of CT values. A volume image is composed of a plurality of voxels arranged three-dimensionally. Each voxel is assigned a CT value.

The medical image diagnostic apparatus 20 transmits the generated CT image data to the image processing apparatus 10 via the hospital network. When receiving the CT image data transmitted from the medical image diagnostic apparatus 20 , the image processing apparatus 10 stores the received CT image data in the storage circuit 15 .

The processing circuit 11 of the image processing apparatus 10 executes the image acquisition function 110 when, for example, an operator inputs an image processing start instruction via the input interface 12 . When the image acquisition function 110 is executed, the processing circuit 11 reads out CT image data desired by the operator from the storage circuit 15 . The processing circuitry 11 executes the likelihood acquisition function 111 when the CT image data is read. When the likelihood acquisition function 111 is executed, the processing circuit 11 acquires a likelihood value for each substance and structure in the body represented by the pixel based on the pixel value (brightness value) of the pixel included in the CT image data. .

Specifically, for example, the processing circuitry 11 acquires likelihood values using a one-dimensional histogram representing frequency distributions of substances and structures in the body. FIG. 4 is a diagram schematically showing an example of processing in which likelihood values are acquired by the likelihood acquisition function 111 shown in FIG. FIG. 4 illustrates an example of calculating a likelihood value for a CT image of a region of interest belonging to the lung field.

The pixel values of the CT image are distributed in a specific range in a histogram set for each substance and structure in the body with reference to -1000 HU for air and 0 HU for water. In particular, it is often distributed in the range of -950 HU or less for air, -950 HU to -850 HU for lung fields (pulmonary parenchyma), -850 HU to -300 HU for ground glass structures, and -300 HU or more for blood vessels. Since the pixel values of the texture pattern also change depending on the type of disease, the range in which the pixel values are distributed is an important feature in identification.

The storage circuit 15 preliminarily stores a histogram as shown in FIG. The processing circuit 11 reads the histogram from the storage circuit 15 and calculates four likelihood values from one pixel value based on the read histogram. Specifically _, the distributions of the four substances shown in FIG. : blood vessels, etc.), the posterior probability can be calculated as follows using Bayes' theorem, where I(x) is the pixel value at the pixel position x.

Here, p(c _k ) represents the prior probability, which may be uniform (=1/4). The processing circuit 11 uses the probability (0 to 1) that the pixel value belongs to each distribution, which is obtained by Equation (1), as the likelihood value. The processing circuit 11 performs, for example, the calculation shown in Equation (1) for all pixels included in the CT image data, and obtains a plurality of likelihood values for each pixel.

Also, for example, the processing circuitry 11 may acquire likelihood values using a two-dimensional co-occurrence histogram. Since the likelihood value in equation (1) is calculated from a single pixel value, it does not contain spatial information. Therefore, it may be difficult to express the shape of the texture pattern. Therefore, a co-occurrence histogram of pixel value pairs at spatially close positions may be used.

FIG. 5 is a diagram schematically showing an example of processing for acquiring likelihood values using co-occurrence histograms. FIG. 5 illustrates an example of calculating the likelihood value for a CT image of a region of interest belonging to the lung field. The storage circuit 15 preliminarily stores co-occurrence histograms as shown in FIG. The processing circuit 11, in the likelihood acquisition function 111, reads the co-occurrence histogram from the storage circuit 15, and calculates a plurality of likelihood values from one pixel value based on the read-out co-occurrence histogram. Specifically, the processing circuit 11 sets the pixel value at the pixel position x as I(x) and the pixel value at the position y, which is a distance r from x, as I(y), and sets the simultaneous probability density function p(I(x ), I(y)|c _k , r), (k=1: air, 2: lung parenchyma, 3: ground glass structure, 4: blood vessels, etc.). Then, the processing circuit 11 calculates the posterior probability as shown in Equation (2), and uses the calculated posterior probability as the likelihood value.

Note that when r=0, x=y, so the posterior probability obtained by equation (2) is the same value as the posterior probability obtained by equation (1). Here, a plurality of candidates can be considered for the pixel position y. For example, as shown in FIG. Good. FIG. 5 shows an example of the results calculated with r as a parameter (r=1 to 5). Four likelihood values can be obtained for each distance r, and when all five r=1 to 5 are used, the total number of obtained likelihood values is 20 (=4×5). The processing circuit 11 performs, for example, the calculation shown in Equation (2) for all pixels included in the CT image data, and obtains a plurality of likelihood values for each pixel.

Using multiple distances makes it possible to capture structures at different scales. Moreover, it becomes possible to identify the shape of the texture pattern while taking spatial information into account. Note that the processing circuit 11 can extend the processing using a two-dimensional co-occurrence histogram to processing using a multi-dimensional co-occurrence histogram using three or more pixel values.

Further, for example, the processing circuit 11 may acquire the likelihood value using decision tree learning. FIG. 7 is a diagram schematically showing an example of processing for acquiring likelihood values using decision tree learning. FIG. 7 illustrates an example of calculating the likelihood value for a CT image of a region of interest belonging to the lung field. According to FIG. 7, the processing circuit 11 determines the likelihood value based on the frequency distribution set at the terminal node of the decision tree (see Non-Patent Document 2, for example).

Specifically, for example, as shown in FIG. 7, a plurality of tree structures are prepared in advance for branching based on the magnitude of the pixel value difference between two random pixels in a local region of a predetermined size. (M) to learn. For example, if four patterns of air, lung parenchyma, frosted glass, and blood vessels are used as input regions, the likelihood of air, lung parenchyma, frosted glass, and blood vessels is calculated by the frequency distribution of learning samples that reach terminal nodes. It is possible to calculate A plurality of widths are set for the local region, and the tree structure is learned for each local region of the set width. Information about tree structure learning is stored in the storage circuit 15 .

In the likelihood acquisition function 111, when a local region of a predetermined size is input, the processing circuit 11 selects a random two-point pixel pair within the region according to a learned decision tree. The processing circuit 11 repeats branching based on the size relationship between the two pixels in the local region, and acquires the frequency distribution of the learning samples set at the terminal node. The processing circuitry 11 calculates, for example, likelihood values for air, lung parenchyma, frosted glass, and blood vessels based on the acquired frequency distributions. For example, the processing circuitry 11 may take the average value of the likelihood values for each of a plurality of (M) trees as the likelihood values for air, lung parenchyma, frosted glass, and blood vessels.

The processing circuit 11 obtains a plurality of likelihood values for all pixels included in the CT image data while sliding the input predetermined local region. In addition, the processing circuit 11 performs similar processing on a plurality of local regions having different sizes stored in the storage circuit 15, and acquires a plurality of likelihood values for each of the stored local regions.

Further, for example, the processing circuit 11 may acquire the likelihood value using an enhancement filter. FIG. 8 is a diagram schematically showing an example of processing for acquiring likelihood values using an enhancement filter. FIG. 8 illustrates an example of calculating the likelihood value for a CT image of a region of interest belonging to the lung field. The lung includes mass structures such as nodules, tubular structures such as blood vessels, and membranous structures such as the mesenchyme. Therefore, obtaining the likelihood values of mass structures, tube structures, and membrane structures is important information in identifying texture patterns. According to FIG. 8, the processing circuit 11 obtains likelihood values based on the substance and structure enhancement filters in the body (see, for example, Non-Patent Document 3).

The processing circuit 11 may have, for example, an identification function using a trained neural network or the like for obtaining likelihood values of structures. At this time, the memory circuit 15 stores, for example, data related to machine learning that has been learned in advance. The processing circuitry 11 uses a trained neural network based on the data stored in the storage circuitry 15 in its identification function to obtain likelihood values for structures.

In the description of the likelihood acquisition function 111 using FIGS. 4 to 8, the processing circuit 11 extracts air, lung parenchyma, ground-glass shadows, blood vessels, mass structures, tube structures, from one pixel included in the CT image. and membrane structures, the case of obtaining likelihood values for seven types of substances and structures has been described as an example. However, it is not limited to this. Processing circuitry 11 may obtain likelihood values for other materials and structures. For example, a CT image may contain solid shadows and nodules. Also, the CT image may contain a line structure instead of a tube structure, and may contain a plate structure instead of a membrane structure. In addition, when the CT image includes regions other than the lung, the CT image includes gas, adipose tissue, water, soft tissue, calcified tissue, etc., which can be classified based on the pixel values. good too. The processing circuitry 11 provides for air, lung parenchyma, ground-glass opacities, solid opacities, blood vessels, nodules, gas, adipose tissue, water, soft tissue, calcified tissue, mass structures, tubular or linear structures, and membrane or plate structures. At least one of the likelihood values may be acquired.

Next, the processing of the feature amount calculation function 112 shown in FIG. 2 will be specifically described. For example, the processing circuit 11 executes the feature amount calculation function 112 after acquiring the likelihood value for each pixel included in the CT image. FIG. 9 is a diagram schematically showing an example of processing in which feature vectors are created by the feature quantity calculation function 112 shown in FIG. In the feature amount calculation function 112, the processing circuit 11 applies an appropriate weighting factor to the N likelihood values acquired for each pixel of the CT image by the likelihood acquisition function 111. FIG. The processing circuit 11 creates one feature vector by summing the weighted likelihood values over the entire image.

Specifically, the processing circuit 11 generates a vector _{w i} in which weight coefficients of the same length as the vector v _i are arranged, assuming that a vector in which the likelihood values of the channel i (i=1 to N) are arranged is a vector v _i K _i pieces of are prepared. The processing circuit 11 arranges the K _i vectors w _i into a matrix W _i shown below.

In equation (3), V represents the number of pixels.

The processing circuit 11 uses the matrix W _i to calculate a vector u in which feature quantities are arranged as follows.

In Equation (4), P represents the total number of features. The processing circuit 11 creates a feature vector by adding the calculated vector u over the entire image.

Note that the processing circuit 11 may calculate the vector u by adding the bias vector b as follows.

Also, in the description using the equations (3) to (5), a weighting factor is set for each channel. However, it is not limited to this. The weighting factors may be the same.

The weighting factors applied to the likelihood values are determined by various techniques. For example, the weighting factors may use fixed values such as Gaussian filters, Gabor filters, mean filters, and box filters. Also, the weighting factor may determine an optimum value by machine learning. FIG. 10 is a diagram schematically showing an example of processing for learning weighting factors using a neural network. For example, a network is prepared in which output units corresponding to the number of texture patterns to be identified in advance and feature amounts obtained by multiplication with weighting factors are connected in a fully connected layer. It should be noted that the initial value of the weighting factor may be set randomly from a Gaussian distribution, a uniform distribution, or the like. Then, the weighting factors are iteratively updated using, for example, error backpropagation. When using machine learning, weighting factors are automatically determined according to the problem to be identified. Therefore, the identification accuracy may be improved compared to using a predetermined weighting factor such as a Gaussian filter.

As described above, in the present embodiment, the processing circuitry 11 acquires desired medical image data using the image acquisition function 110 . The processing circuit 11 uses the likelihood acquisition function 111 to acquire, for each pixel of the medical image, likelihood values relating to multiple types of substances and structures in the body according to the pixel value. Then, the processing circuit 11 uses the feature amount calculation function 112 to calculate a plurality of feature amounts using the acquired plurality of likelihood values. Thus, processing circuitry 11 is adapted to perform feature extraction using the likelihood of materials and structures within the body associated with the texture pattern to be identified, rather than extracting features directly from the image. As a result, the processing circuit 11 can obtain a feature amount with high discrimination ability.

Therefore, according to the image processing apparatus 10 of the present embodiment, it is possible to improve the accuracy of identifying the texture pattern of the medical image.

In the above-described embodiment, the case where the processing circuit 11 has the image acquisition function 110, the likelihood acquisition function 111, and the feature amount calculation function 112 has been described as an example. However, it is not limited to this. For example, as shown in FIG. 11, processing circuitry 11a may include an identification feature 113. FIG. The processing circuit 11 a has an image acquisition function 110 , a likelihood acquisition function 111 , a feature quantity calculation function 112 , and an identification function 113 by executing programs stored in the storage circuit 15 .

The identification function 113 uses machine learning learned using the feature vectors acquired by the feature amount calculation function 112 to identify preset texture patterns such as lesions, body tissues, or organs. At this time, for example, the memory circuit 15 stores in advance data related to machine learning for identifying a predetermined texture pattern. Although the number of patterns identified by the identification function 113 is arbitrary, it may be more or less than the types of likelihood values obtained. The number of identified patterns is, for example, two when distinguishing between benign and malignant. FIG. 12 is a diagram schematically explaining the operation of the processing circuit 11a shown in FIG. According to FIG. 12, the processing circuit 11a generates likelihood images of likelihood 1 to likelihood N by obtaining a likelihood value for each pixel of the medical image. The processing circuit 11a generates a feature vector based on likelihood images of likelihood 1 to likelihood N. FIG. Then, the processing circuit 11a uses machine learning learned using the feature vectors acquired as samples, and identifies preset texture patterns such as lesions, body tissues, or organs based on the acquired feature vectors.

Machine learning algorithms used in the identification function 113 include discriminant analysis, logistic regression, support vector machines, neural networks, randomized trees, subspace methods, and the like. By combining any of these with the conditional probability field and the graph cut, the texture pattern may be identified in consideration of the adjacency of pixels.

Further, in the present embodiment, the case where the image acquisition function 110, the likelihood acquisition function 111, the feature amount calculation function 112, and the identification function 113 are provided in the image processing apparatus 10 has been described as an example. However, it is not limited to this. The image acquisition function 110 , the likelihood acquisition function 111 , the feature quantity calculation function 112 , and the identification function 113 may be provided in the medical image diagnostic apparatus 20 . FIG. 13 shows the functions of an X-ray CT apparatus in the case where the X-ray CT apparatus, which is an example of the medical image diagnostic apparatus 20, has an image acquisition function 110, a likelihood acquisition function 111, a feature value calculation function 112, and an identification function 113. It is a figure which shows the example of a structure. 13, the processing circuit 21 of the medical image diagnostic apparatus 20 executes programs stored in the storage circuit 22 to perform an image acquisition function 110, a likelihood acquisition function 111, a feature amount calculation function 112, and It implements the identification function 113 . Note that the processing circuit 21 may perform the likelihood acquisition function 111 on CT image data generated by image reconstruction processing. In this case, processing circuitry 21 need not necessarily implement image acquisition function 110 .

The image acquisition function 110, the likelihood acquisition function 111, the feature amount calculation function 112, and the identification function 113 according to the present embodiment can be realized by using, for example, a general-purpose computer device as basic hardware. That is, the image processing apparatus 10 and the medical image diagnostic apparatus 20 have an image acquisition function 110, a likelihood acquisition function 111, a feature amount calculation function 112, and an identification function 113 by causing a processor installed in a computer device to execute a program. can be realized. At this time, the image processing apparatus 10 and the medical image diagnostic apparatus 20 implement the image acquisition function 110, the likelihood acquisition function 111, the feature value calculation function 112, and the identification function 113 by installing the above programs in advance. good too. In addition, the image processing apparatus 10 and the medical image diagnostic apparatus 20 read the above program from a storage medium such as a CD-ROM, or receive it via a network and install it as appropriate, so that the image acquisition function 110, the likelihood The acquisition function 111, the feature amount calculation function 112, and the identification function 113 may be realized. Also, the three-dimensional CT image input to the image processing apparatus 10 and the positional information specifying the structure to be extracted can be stored in the memory, hard disk, or CD-R or CD-RW that is built-in or external to the computer device. , DVD-RAM, and DVD-R.

Further, in the above embodiments, a CT image was given as an example of a medical image. However, medical images are not limited to CT images. Other images such as MR images acquired by an MRI apparatus may be used. The processing circuitry 11 obtains likelihood values based on, for example, structures contained in the MR image. Then, the processing circuit 11 calculates a feature amount by multiplying the obtained likelihood value by a weighting factor, and obtains a feature vector.

The term "processor" used in the above description is, for example, a CPU (central processing unit), a GPU (Graphics Processing Unit), or an application specific integrated circuit (ASIC), a programmable logic device (e.g., Circuits such as Simple Programmable Logic Device (SPLD), Complex Programmable Logic Device (CPLD), Field Programmable Gate Array (FPGA), etc. A processor is a memory circuit. The functions are realized by reading and executing the program stored in the memory circuit 15. Instead of storing the program in the storage circuit 15, the program may be directly incorporated in the circuit of the processor. A processor realizes its functions by reading and executing a program embedded in a circuit.The processors described in the above embodiments are not limited to being configured as a single circuit, and may be composed of a plurality of independent circuits. They may be combined into one processor to realize its function.

The image acquisition function 110, the likelihood acquisition function 111, the feature quantity calculation function 112, and the identification function 113 in this embodiment are implemented by the corresponding image acquisition unit, likelihood acquisition unit, feature quantity calculation unit, and identification unit, respectively. can be anything. It should be noted that the components described as "units" in the present embodiment may be implemented by hardware or software, or may be implemented by hardware and software. It may be realized by a combination of

While several embodiments of the invention have been described, these embodiments have been presented by way of example and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, replacements, and modifications can be made without departing from the scope of the invention. These embodiments and their modifications are included in the scope and spirit of the invention, as well as the scope of the invention described in the claims and equivalents thereof.

DESCRIPTION OF SYMBOLS 1... Medical information system 10... Image processing apparatus 11, 11a... Processing circuit 110... Image acquisition function 111... Likelihood acquisition function 112... Feature-value calculation function 113... Identification function 12... Input interface 13 ... output interface, 14 ... communication interface, 15 ... storage circuit, 20 ... medical image diagnostic apparatus, 21 ... processing circuit, 22 ... storage circuit, 30 ... image storage device.

Claims (16)

an image acquisition unit that acquires medical image data;
a plurality of likelihood values representing the likelihood that an element included in a medical image based on the medical image data is classified into at least one of classification items corresponding to substances and structures in the body, for each pixel of the medical image; a likelihood acquisition unit to acquire;
A feature amount calculation unit that calculates a plurality of feature amounts using the plurality of likelihood values;
an identification unit that identifies regions of lesions, body tissues, or organs using the plurality of feature quantities;
and
The identification unit uses the plurality of feature amounts to identify a texture pattern corresponding to a lesion, body tissue, or organ,
the substances and structures in the body are substances or structures contained in the texture pattern to be identified;
The likelihood acquisition unit acquires the likelihood value for each pixel according to the pixel value of the pixel and the pixel values of pixels surrounding the pixel.
Image processing device. The likelihood acquisition unit sets a plurality of distances between each pixel and pixels surrounding the pixel, and for each distance, 2. The image processing apparatus according to claim 1, wherein said likelihood value is acquired. 3. The image processing apparatus according to claim 1, wherein the feature amount calculation unit calculates the plurality of feature amounts by multiplying the plurality of likelihood values by preset weighting factors. 3. The image processing apparatus according to claim 1, wherein the likelihood acquisition unit acquires the likelihood value based on a frequency distribution of substances and structures in the body corresponding to the pixel value. 5. The image processing apparatus according to claim 4 , wherein said frequency distribution includes a multidimensional co-occurrence histogram. The image processing apparatus according to any one of claims 1 to 5, wherein the medical image includes a CT image. 7. The image processing apparatus according to claim 6, wherein the substance in the body includes at least one of air, gas, fatty tissue, water, soft tissue, and calcified tissue, which can be classified based on pixel values of CT images. 8. The image processing apparatus according to any one of claims 1 to 7, wherein the intracorporeal structure includes at least one of a body tissue mass structure, a tube or line structure, and a plate or membrane structure. 7. The image processing apparatus according to claim 6, wherein said medical image includes an image of lungs. 10. The image processing apparatus according to claim 9, wherein the substance in the body includes at least one of air, lung parenchyma, ground-glass opacity, solid opacity, blood vessel, and nodule. 2. The image processing apparatus according to claim 1, wherein said likelihood acquisition unit calculates said likelihood value based on a co-occurrence histogram. 2. The image processing apparatus according to claim 1, wherein said likelihood acquisition unit calculates said likelihood value using decision tree learning. 2. The image processing apparatus according to claim 1, wherein said likelihood acquisition unit acquires said likelihood value based on a trained neural network. The feature amount calculation unit calculates the feature amount based on a network that connects output units corresponding to the number of texture patterns to be identified in advance and the feature amount obtained by multiplying the weight coefficient with a fully connected layer. 2. The image processing apparatus according to claim 1, wherein The identifying unit uses a machine learning algorithm to identify the type of lesion, body tissue, or organ included in the medical image,
2. The image processing apparatus according to claim 1, wherein said machine learning algorithm is at least one of discriminant analysis, logistic regression, support vector machine, neural network, randomized trees, subspace method. The identifying unit uses a machine learning algorithm to identify the type of lesion, body tissue, or organ included in the medical image,
2. The machine learning algorithm of claim 1, wherein the machine learning algorithm combines any of discriminant analysis, logistic regression, support vector machines, neural networks, randomized trees, and subspace methods with conditional random fields or graph cuts. image processing device.