WO2019201204A1

WO2019201204A1 - Method and device for detecting nodule in image

Info

Publication number: WO2019201204A1
Application number: PCT/CN2019/082677
Authority: WO
Inventors: 魏子昆; 杨忠程; 华铱炜
Original assignee: 杭州依图医疗技术有限公司; 杭州依图网络科技有限公司; 广州依图医疗技术有限公司
Priority date: 2018-04-17
Filing date: 2019-04-15
Publication date: 2019-10-24
Also published as: CN108648178A

Abstract

A method and a device for detecting a nodule in an image. The method comprises: acquiring a nodule image and three-dimensional coordinates of candidate nodules in the nodule image (201); determining, according to the three-dimensional coordinates of the candidate nodules, an ROI containing the candidate nodules from the nodule image (202); determining a degree of confidence of the candidate nodules according to the ROI and a nodule detection model (203); and filtering out a false-positive candidate nodule in the candidate nodules having a degree of confidence greater than a threshold according to candidate nodules having a degree of confidence greater than a threshold, a segmentation result of a body part in which the candidate nodules are located, and the three-dimensional coordinates of the candidate nodules, so as to determine a nodule in the nodule image and a degree of confidence corresponding to the nodule (204). A convolutional neural network is used to conduct training on nodule images in which a nodule region has been marked, to obtain a nodule detection model, and an ROI is input into the nodule detection model to obtain a degree of confidence of candidate nodules and to improve the efficiency of nodule detection. Further, false-positive nodules are filtered out after nodules are detected, thereby improving the accuracy of nodule detection.

Description

Method and device for detecting image nodules

The present application claims priority to Chinese Patent Application No. 20 181 034 529 7.3 filed on Apr. 17, 2018, the entire disclosure of which is incorporated herein by reference. In this application.

Technical field

Embodiments of the present invention relate to the field of machine learning technologies, and in particular, to a method and an apparatus for detecting image nodules.

Background technique

At present, nodules have caused widespread concern, such as pulmonary nodules, thyroid nodules, etc., doctors generally observe the patient's nodules through medical imaging. As the patient's nodules may change over time, for example, increase, decrease, or grow new nodules and the like. At present, the way to find nodules from medical images is mainly to find the corresponding nodules in the medical images by doctors manually observing the medical images, but this situation will lead to the accuracy of the found nodules, and also It takes a lot of time and is subjective.

Summary of the invention

Embodiments of the present invention provide a method and apparatus for detecting image nodules, which are used to improve the efficiency of nodule detection and improve the accuracy of nodule detection.

A method for detecting image nodules according to an embodiment of the present invention includes:

Obtaining a nodule image and three-dimensional coordinates of candidate nodules in the nodule image;

Determining a Region of Interest (ROI) containing the candidate nodule from the nodule image according to the three-dimensional coordinates of the candidate nodule;

Determining a confidence level of the candidate nodule according to the ROI and a nodule detection model, wherein the nodule detection model is determined by using a convolutional neural network to train a plurality of nodule effects of the labeled nodule region;

Filtering the candidate nodules of the false positives in the candidate nodules whose confidence is greater than the threshold according to the candidate nodules whose confidence is greater than the threshold, the segmentation result of the body part where the candidate nodule is located, and the three-dimensional coordinates of the candidate nodules, A nodule in the nodule image and a confidence level corresponding to the nodule are determined.

Since the convolutional neural network is used to train the nodule image of the marked nodule region to obtain the nodule detection model, the ROI can be directly input to the nodule detection model to obtain the confidence of the corresponding candidate nodule, and the improved knot The efficiency of the section detection, and the false-positive nodules are filtered out after the nodules are detected, thereby improving the accuracy of the nodule detection.

Optionally, determining, according to the three-dimensional coordinates of the candidate nodule, the ROI of the candidate nodule from the nodule image, including:

Centering on the three-dimensional coordinates of the candidate nodule, extending a preset distance to the periphery to determine a pixel cube including the candidate nodule, the preset distance being a preset multiple of a radius of the candidate nodule;

A spatial information channel is added to each pixel of the pixel cube to determine the ROI, and the spatial information channel is a distance between the pixel and the three-dimensional coordinates of the nodule.

The input value input to the nodule detection model can be obtained by the determined ROI, so that the input of the nodule detection model is applied, and the uniformity of the input values is improved.

Optionally, the determining, according to the ROI and the nodule detection model, the confidence that the candidate nodule is included, including:

The ROI is sequentially extracted by the M 3D convolution feature extraction modules to extract the feature image of the ROI, where M is greater than 0;

The extracted feature image of the ROI is input to the fully connected module, and the confidence of the candidate nodule is determined.

The confidence of the candidate nodule is detected by the nodule detection model, thereby improving the efficiency of nodule detection.

Optionally, the candidate nodule according to the confidence level greater than the threshold, the segmentation result of the body part where the candidate nodule is located, and the three-dimensional coordinates and radius of the candidate nodule are filtered out, and the candidate nodule with the confidence greater than the threshold is filtered out. Candidate nodules for false positives, including:

Filtering out candidate nodules of skeletal false positives in the candidate nodules whose confidence is greater than a threshold according to the three-dimensional coordinates of the candidate nodule and the pixels in the preset region where the candidate nodule is located; or

Filtering off the candidate nodules of the diaphragmatic pseudo-positive in the candidate nodules with the confidence greater than the threshold according to the three-dimensional coordinates and radius of the candidate nodule, and the segmentation result of the body part where the candidate nodule is located; or

According to the three-dimensional coordinates of the candidate nodule and the segmentation result of the body part where the candidate nodule is located, the candidate nodule of the mediastinal false positive in the candidate nodule with the confidence greater than the threshold is filtered out.

By filtering out false positive nodules, the accuracy of nodule detection can be improved.

Optionally, the using the convolutional neural network to train the plurality of nodule effects of the labeled nodule region to determine the nodule detection model comprises:

Obtaining a candidate nodule set to be filtered out of the false positive result and a determination result of the doctor for each candidate nodule in the candidate nodule set;

Data enhancement is performed on candidate nodules in the candidate nodule set to obtain an enhanced candidate nodule set;

Determining an ROI of each candidate nodule in the enhanced candidate nodule set from the nodule image according to the enhanced candidate nodule set and the three-dimensional coordinates of each candidate nodule;

The ROI of each candidate nodule in the enhanced candidate nodule set is trained by a preset 3D convolutional neural network model to obtain the nodule detection model.

Correspondingly, the embodiment of the invention further provides an apparatus for detecting image nodules, comprising:

An acquiring unit, configured to acquire a nodule image and three-dimensional coordinates of candidate nodules in the nodule image;

a determining unit, configured to determine, from the nodule image, an ROI including the candidate nodule according to the three-dimensional coordinates of the candidate nodule;

a detecting unit, configured to determine a confidence level of the candidate nodule according to the ROI and a nodule detection model, wherein the nodule detection model uses a convolutional neural network to perform influence on multiple nodules of the labeled nodule region Determined after training;

a filtering unit, configured to filter out false positives in the candidate nodules whose confidence is greater than a threshold according to a candidate nodule with a confidence greater than a threshold, a segmentation result of a body part where the candidate nodule is located, and a three-dimensional coordinate of the candidate nodule Candidate nodules determine the nodules in the nodule image and the confidence level corresponding to the nodules.

Optionally, the determining unit is specifically configured to:

Optionally, the detecting unit is specifically configured to:

Optionally, the filtering unit is specifically configured to:

Optionally, the detecting unit is specifically configured to:

Correspondingly, an embodiment of the present invention further provides a computer device, including:

a memory for storing program instructions;

And a processor, configured to invoke a program instruction stored in the memory, and execute the method for detecting the image nodule according to the obtained program.

Correspondingly, the embodiment of the present invention further provides a computer readable non-volatile storage medium, comprising computer readable instructions, when the computer device reads and executes the computer readable instructions, causing the computer device to execute the image image Method of section detection.

Correspondingly, an embodiment of the present invention further provides a computer program product, the computer program product comprising a computer program stored on a computer readable non-volatile storage medium, the computer program comprising program instructions, when the program When the instructions are executed by the computer device, the computer device is caused to perform the method of image nodule detection described above.

In the embodiment of the present invention, the convolutional neural network is used to train the nodule image of the marked nodule region to obtain a nodule detection model, thereby directly inputting the ROI into the nodule detection model to obtain a confidence of the corresponding candidate nodule. Degree, improved efficiency of nodule detection, and filtering out false positive nodules after detection of nodules, thereby improving the accuracy of nodule detection.

DRAWINGS

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly described below. It is obvious that the drawings in the following description are only some embodiments of the present invention, Those skilled in the art can also obtain other drawings based on these drawings without paying any creative work.

FIG. 1 is a schematic diagram of a system architecture according to an embodiment of the present invention;

2 is a schematic flowchart of a method for detecting image nodules according to an embodiment of the present invention;

3 is a schematic diagram of a nodule image according to an embodiment of the present invention;

4 is a schematic flowchart of an apparatus for detecting image nodules according to an embodiment of the present invention;

FIG. 5 is a schematic structural diagram of a computer device according to an embodiment of the present invention.

detailed description

The present invention will be further described in detail with reference to the accompanying drawings, in which FIG. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments of the present invention without creative efforts are within the scope of the present invention.

FIG. 1 is a system architecture applicable to a method for detecting image nodules according to an embodiment of the present invention. Referring to FIG. 1, the system architecture can be server 100, including processor 110, communication interface 120, and memory 130.

The communication interface 120 is used for communication by a terminal device applied by a doctor, and transmits and receives information transmitted by the terminal device to implement communication.

The processor 110 is a control center of the server 100 that connects various portions of the entire server 100 using various interfaces and lines, by running or executing software programs and/or modules stored in the memory 130, and recalling data stored in the memory 130, Various functions and processing data of the server 100 are executed. Alternatively, processor 110 may include one or more processing units.

The memory 130 can be used to store software programs and modules, and the processor 110 executes various functional applications and data processing by running software programs and modules stored in the memory 130. The memory 130 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application required for at least one function, and the like; the storage data area may store data created according to business processing and the like. Moreover, memory 130 can include high speed random access memory, and can also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid state storage device.

It should be noted that the structure shown in FIG. 1 is only an example, which is not limited by the embodiment of the present invention.

Based on the above description, FIG. 2 exemplarily shows a flow of image nodule detection provided by an embodiment of the present invention, which may be performed by a device for detecting image nodules.

As shown in FIG. 2, the process specifically includes:

Step 201: Acquire a nodule image and three-dimensional coordinates of candidate nodules in the nodule image.

The nodule image is a three-dimensional image, and the three-dimensional coordinates of the candidate nodule may be the three-dimensional coordinates of the point within the candidate nodule (such as the three-dimensional coordinates of the nodule center point), or may be the three-dimensional coordinates of the point of the candidate nodule surface. Candidate nodules include, but are not limited to, pulmonary nodules, thyroid nodules, and breast nodules. The nodule image may be a Computed Tomography (CT) image, a Magnetic Resonance Imaging (MRI) image, or the like. To more clearly describe the nodule image, FIG. 3 exemplarily shows CT images of the lungs of a patient.

Step 202: Determine an ROI of the candidate nodule from the nodule image according to the three-dimensional coordinates of the candidate nodule.

Specifically, the preset distance may be extended to the periphery by using the three-dimensional coordinates of the candidate nodule as a center, and the pixel cube including the candidate nodule is determined, wherein the preset distance is a preset multiple of the radius of the candidate nodule, such as a candidate. The radius of the nodule is 1.25 times. Then intercept the pixel cube and scale it to a certain size. Then, a spatial information channel is added to each pixel in the pixel cube to output an ROI, and the spatial information channel is the distance between the pixel cube and the three-dimensional coordinates of the candidate nodule. For example, it is possible to select a pixel cube of 2L*2L*2L size by extending L pixels in each direction of the three coordinate axes centering on the three-dimensional coordinates of the candidate nodules.

Step 203: Determine a confidence level of the candidate nodule according to the ROI and the nodule detection model.

In the embodiment of the present invention, the nodule detection model is obtained by using a convolutional neural network to train a plurality of nodules in the marked nodule region, and specifically: obtaining a candidate for filtering a false positive result first. The nodule set and the doctor determine the candidate nodules in the candidate nodule set, and the candidate nodule set of the false positive result to be filtered is obtained by collecting a large number of chest CT images and using other programs, and then passing through multiple doctors. The candidate nodules in the candidate nodule set are determined to determine whether it is a nodule.

The candidate nodules in the candidate nodule set are then data enhanced to obtain an enhanced candidate nodule set. For example, the amount of data can be increased to the previous K times. The possible way is to enhance the data amount by random horizontal mirroring, random rotation of arbitrary angles, random up and down and left and right translation of 0 to 5 pixels, and randomization of 0.85 to 1.15 times. K times before.

The ROI of each candidate nodule in the enhanced candidate nodule set is determined from the nodule image according to the enhanced candidate nodule set and the three-dimensional coordinates of each candidate nodule. For the specific determination method, refer to step 203. Narration.

Finally, the ROI of each candidate nodule in the enhanced candidate nodule set is trained by a preset 3D convolutional neural network model to obtain a nodule detection model. During training, the nod confidence of the 3D convolutional neural network model output and the label of the training sample can be cross-entropy as a loss function, and trained by the back propagation method. The training optimization algorithm is SGD.

The nodule detection model obtained by the above steps includes M 3D convolution feature extraction models and a fully connected module. Each 3D convolution feature extraction model also includes a 3D convolutional layer of J*J*J and a max_pool layer of H*H*H. A fully connected model can include two fully connected layers.

When the nodule detection model is specifically used, the ROI obtained in the above step 202 may be sequentially extracted by the M 3D convolution feature extraction module to extract the feature image of the ROI, and then the extracted ROI feature image is input to the fully connected module to determine the candidate. The confidence of the nodule, thereby obtaining the corresponding candidate nodule, improves the efficiency of candidate nodule detection.

Step 204: Filter out candidate for false positive in the candidate nodule with the confidence greater than the threshold according to the candidate nodule with the confidence greater than the threshold, the segmentation result of the body part where the candidate nodule is located, and the three-dimensional coordinate of the candidate nodule Nodule.

In order to obtain a more accurate candidate nodule, it is also necessary to filter and filter the candidate nodules obtained in step 203. Among them, the segmentation result of the body part where the candidate nodule is located is obtained by other means. The body part may be a body part such as a lung, a thyroid gland, a breast, or the like, which is not limited in the embodiment of the present invention.

During the specific filtering process, the candidate nodules of the skeleton-like false positives in the candidate nodules whose confidence is greater than the threshold may be filtered according to the three-dimensional coordinates of the candidate nodules and the pixels in the preset region where the candidate nodules are located. This threshold can be set empirically. For example, three-dimensional coordinates from the beginning of the nodule candidate, Imm extended to the surrounding area, a total area I * I * Imm ³ a. Then, the pixels with CT values greater than 400 in the statistical region can be regarded as candidate nodules of false positives of the bones if they occupy a larger proportion than the first threshold, so that they can be filtered out.

The candidate nodules of the diaphragmatic false positives in the candidate nodules with confidence greater than the threshold may also be filtered according to the three-dimensional coordinates and radius of the candidate nodules and the segmentation results of the body parts where the candidate nodules are located. For example, starting from the three-dimensional coordinates of the candidate nodule, the area block of the diameter is expanded to the periphery, and the number of pixels in the lung and the lung outside the block is counted, if the number of pixels in the lung and the number of pixels outside the lung are in the nodule image. The proportions occupied are similar, and basically in the middle of the image, it can be regarded as a candidate nodule of the diaphragmatic false positive, which can be filtered out.

Candidate nodules of mediastinal false positives in candidate nodules with confidence greater than the threshold can also be filtered out according to the three-dimensional coordinates of the candidate nodules and the segmentation results of the body parts where the candidate nodules are located. For example, if the center of the candidate nodule is outside the lung, and the vertical direction does not exceed the lung range, and the central position, it can be said that the relative position on the X-axis is between 0.45 and 0.55, then it can be regarded as a mediastinum. Positive candidate nodule.

Through the above filtering method, after filtering out the false positive candidate nodules, the confidence of the final nodule is obtained, so that the accuracy of the nodule detection can be improved.

The above embodiment shows that by acquiring the nodule image and the three-dimensional coordinates of the candidate nodule in the nodule image, the ROI containing the candidate nodule is determined from the nodule image according to the three-dimensional coordinates of the candidate nodule, and is determined according to the ROI and the nodule detection model. Based on the confidence of the candidate nodule, the nodule detection model is determined by using a convolutional neural network to train the effects of multiple nodules in the labeled nodule region. According to the candidate nodule with the confidence greater than the threshold, the candidate nodule is located. The segmentation result of the body part and the three-dimensional coordinates of the candidate nodule, filtering out the candidate nodule of the false positive in the candidate nodule with the confidence greater than the threshold, determining the nodule in the nodule image and the confidence corresponding to the nodule . Since the convolutional neural network is used to train the nodule image of the marked nodule region to obtain the nodule detection model, the ROI can be directly input to the nodule detection model to obtain the confidence of the corresponding candidate nodule, and the improved knot The efficiency of the section detection, and the false-positive nodules are filtered out after the nodules are detected, thereby improving the accuracy of the nodule detection.

Based on the same technical concept, FIG. 4 exemplarily shows an apparatus for detecting image nodules according to an embodiment of the present invention, which can perform a process of image nodule detection.

As shown in FIG. 4, the device specifically includes:

An obtaining unit 401, configured to acquire a nodule image and three-dimensional coordinates of candidate nodules in the nodule image;

a determining unit 402, configured to determine, according to the three-dimensional coordinates of the candidate nodule, an ROI including the candidate nodule from the nodule image;

a detecting unit 403, configured to determine a confidence level of the candidate nodule according to the ROI and a nodule detection model, wherein the nodule detection model uses a convolutional neural network to affect multiple nodules of the labeled nodule region Determined after training;

The filtering unit 404 is configured to filter out the false candidate in the candidate nodule with the confidence greater than the threshold according to the candidate nodule with the confidence greater than the threshold, the segmentation result of the body part where the candidate nod is located, and the three-dimensional coordinate of the candidate nodule A positive candidate nodule determines the nodule in the nodule image and the confidence level corresponding to the nodule.

Optionally, the determining unit 402 is specifically configured to:

Optionally, the detecting unit 403 is specifically configured to:

Optionally, the filtering unit 404 is specifically configured to:

Optionally, the detecting unit 403 is specifically configured to:

Based on the same technical concept, an embodiment of the present invention further provides a computer device, as shown in FIG. 5, including:

a memory 501, configured to store program instructions;

The processor 502 is configured to invoke a program instruction stored in the memory 501, and execute the method for detecting the image nodule according to the obtained program.

Based on the same technical concept, an embodiment of the present invention further provides a computer readable non-volatile storage medium, comprising computer readable instructions, when the computer device reads and executes the computer readable instructions, causing the computer device to execute The above method of image nodule detection.

Based on the same technical concept, an embodiment of the present invention further provides a computer program product, the computer program product comprising a computer program stored on a computer readable non-volatile storage medium, the computer program comprising program instructions The method of causing the computer device to perform the image nodule detection described above when the program instructions are executed by a computer device.

The present invention has been described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (system), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or FIG. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing device to produce a machine for the execution of instructions for execution by a processor of a computer or other programmable data processing device. Means for implementing the functions specified in one or more of the flow or in a block or blocks of the flow chart.

The computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture comprising the instruction device. The apparatus implements the functions specified in one or more blocks of a flow or a flow and/or block diagram of the flowchart.

These computer program instructions can also be loaded onto a computer or other programmable data processing device such that a series of operational steps are performed on a computer or other programmable device to produce computer-implemented processing for execution on a computer or other programmable device. The instructions provide steps for implementing the functions specified in one or more of the flow or in a block or blocks of a flow diagram.

While the preferred embodiment of the invention has been described, it will be understood that Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and the modifications and

It is apparent that those skilled in the art can make various modifications and variations to the invention without departing from the spirit and scope of the invention. Thus, it is intended that the present invention cover the modifications and modifications of the invention

Claims

A method for detecting image nodules, comprising:

Obtaining a nodule image and three-dimensional coordinates of candidate nodules in the nodule image;

Determining a region of interest ROI including the candidate nodule from the nodule image according to the three-dimensional coordinates of the candidate nodule;

Determining a confidence level of the candidate nodule according to the ROI and a nodule detection model, wherein the nodule detection model is determined by using a convolutional neural network to train a plurality of nodule effects of the labeled nodule region;

Filtering the candidate nodules of the false positives in the candidate nodules whose confidence is greater than the threshold according to the candidate nodules whose confidence is greater than the threshold, the segmentation result of the body part where the candidate nodule is located, and the three-dimensional coordinates of the candidate nodules, A nodule in the nodule image and a confidence level corresponding to the nodule are determined.
The method of claim 1, wherein the determining, based on the three-dimensional coordinates of the candidate nodule, the ROI including the candidate nodule from the nodule image comprises:

Centering on the three-dimensional coordinates of the candidate nodule, extending a preset distance to the periphery to determine a pixel cube including the candidate nodule, the preset distance being a preset multiple of a radius of the candidate nodule;

A spatial information channel is added to each pixel of the pixel cube to determine the ROI, and the spatial information channel is a distance between the pixel and the three-dimensional coordinates of the nodule.
The method of claim 1, wherein the determining the confidence of the candidate nodule based on the ROI and the nodule detection model comprises:

The ROI is sequentially extracted by the M 3D convolution feature extraction modules to extract the feature image of the ROI, where M is greater than 0;

The extracted feature image of the ROI is input to the fully connected module, and the confidence of the candidate nodule is determined.
The method according to claim 1, wherein said filtering is performed according to a candidate nodule having a confidence greater than a threshold, a segmentation result of a body part where the candidate nodule is located, and a three-dimensional coordinate and radius of said candidate nodule Candidate nodules for false positives in candidate nodules with confidence greater than the threshold, including:

Filtering out candidate nodules of skeletal false positives in the candidate nodules whose confidence is greater than a threshold according to the three-dimensional coordinates of the candidate nodule and the pixels in the preset region where the candidate nodule is located; or

Filtering off the candidate nodules of the diaphragmatic pseudo-positive in the candidate nodules with the confidence greater than the threshold according to the three-dimensional coordinates and radius of the candidate nodule, and the segmentation result of the body part where the candidate nodule is located; or

According to the three-dimensional coordinates of the candidate nodule and the segmentation result of the body part where the candidate nodule is located, the candidate nodule of the mediastinal false positive in the candidate nodule with the confidence greater than the threshold is filtered out.
The method of claim 1 wherein said convolutional neural network is used to train a plurality of nodule effects of the labeled nodule region to determine said nodule detection model, comprising:

Obtaining a candidate nodule set to be filtered out of the false positive result and a determination result of the doctor for each candidate nodule in the candidate nodule set;

Data enhancement is performed on candidate nodules in the candidate nodule set to obtain an enhanced candidate nodule set;

Determining an ROI of each candidate nodule in the enhanced candidate nodule set from the nodule image according to the enhanced candidate nodule set and the three-dimensional coordinates of each candidate nodule;

The ROI of each candidate nodule in the enhanced candidate nodule set is trained by a preset 3D convolutional neural network model to obtain the nodule detection model.
An apparatus for detecting image nodules, comprising:

An acquiring unit, configured to acquire a nodule image and three-dimensional coordinates of candidate nodules in the nodule image;

a determining unit, configured to determine, from the nodule image, a region of interest ROI including the candidate nodule according to the three-dimensional coordinates of the candidate nodule;

a detecting unit, configured to determine a confidence level of the candidate nodule according to the ROI and a nodule detection model, wherein the nodule detection model uses a convolutional neural network to perform influence on multiple nodules of the labeled nodule region Determined after training;

a filtering unit, configured to filter out false positives in the candidate nodules whose confidence is greater than a threshold according to a candidate nodule with a confidence greater than a threshold, a segmentation result of a body part where the candidate nodule is located, and a three-dimensional coordinate of the candidate nodule Candidate nodules determine the nodules in the nodule image and the confidence level corresponding to the nodules.
The device according to claim 6, wherein the determining unit is specifically configured to:

Centering on the three-dimensional coordinates of the candidate nodule, extending a preset distance to the periphery to determine a pixel cube including the candidate nodule, the preset distance being a preset multiple of a radius of the candidate nodule;

A spatial information channel is added to each pixel of the pixel cube to determine the ROI, and the spatial information channel is a distance between the pixel and the three-dimensional coordinates of the nodule.
The device according to claim 6, wherein the detecting unit is specifically configured to:

The ROI is sequentially extracted by the M 3D convolution feature extraction modules to extract the feature image of the ROI, where M is greater than 0;

The extracted feature image of the ROI is input to the fully connected module, and the confidence of the candidate nodule is determined.
The device according to claim 6, wherein the filtering unit is specifically configured to:

Filtering out candidate nodules of skeletal false positives in the candidate nodules whose confidence is greater than a threshold according to the three-dimensional coordinates of the candidate nodule and the pixels in the preset region where the candidate nodule is located; or

Filtering off the candidate nodules of the diaphragmatic pseudo-positive in the candidate nodules with the confidence greater than the threshold according to the three-dimensional coordinates and radius of the candidate nodule, and the segmentation result of the body part where the candidate nodule is located; or

According to the three-dimensional coordinates of the candidate nodule and the segmentation result of the body part where the candidate nodule is located, the candidate nodule of the mediastinal false positive in the candidate nodule with the confidence greater than the threshold is filtered out.
The device according to claim 6, wherein the detecting unit is specifically configured to:

Obtaining a candidate nodule set to be filtered out of the false positive result and a determination result of the doctor for each candidate nodule in the candidate nodule set;

Data enhancement is performed on candidate nodules in the candidate nodule set to obtain an enhanced candidate nodule set;

Determining an ROI of each candidate nodule in the enhanced candidate nodule set from the nodule image according to the enhanced candidate nodule set and the three-dimensional coordinates of each candidate nodule;

The ROI of each candidate nodule in the enhanced candidate nodule set is trained by a preset 3D convolutional neural network model to obtain the nodule detection model.
A computer device, comprising:

a memory for storing program instructions;

And a processor, configured to invoke the program instructions stored in the memory, and execute the method according to any one of claims 1 to 5 according to the obtained program.
A computer readable non-volatile storage medium, comprising computer readable instructions for causing a computer device to perform any one of claims 1 to 5 when the computer device reads and executes the computer readable instructions Said method.
A computer program product, comprising: a computer program stored on a computer readable non-volatile storage medium, the computer program comprising program instructions, when the program instructions are executed by a computer device, causing the computer The apparatus performs the method of any one of claims 1 to 5.