CN112288752B

CN112288752B - Full-automatic coronary calcified focus segmentation method based on chest flat scan CT

Info

Publication number: CN112288752B
Application number: CN202011181703.0A
Authority: CN
Inventors: 王怡宁; 金征宇; 王健; 徐橙; 郭恒; 许敏丰; 迟颖
Original assignee: Peking Union Medical College Hospital Chinese Academy of Medical Sciences
Current assignee: Peking Union Medical College Hospital Chinese Academy of Medical Sciences
Priority date: 2020-10-29
Filing date: 2020-10-29
Publication date: 2021-08-27
Anticipated expiration: 2040-10-29
Also published as: LU500798A1; LU500798B1; CN112288752A

Abstract

The invention discloses a full-automatic segmentation method of coronary calcified foci based on chest flat scan CT, which comprises the following steps: preprocessing an original chest flat scan CT image to obtain preprocessed data; inputting the preprocessed data into a heart segmentation model, segmenting a heart region in the original chest flat-scan CT image, and obtaining a heart segmentation image which comprises the heart region and corresponds to the original chest flat-scan CT image; screening suspected coronary calcifications in the heart area; extracting a plurality of candidate sample blocks from the screened suspected coronary calcifications; and obtaining coronary artery calcification focus segmentation results of the candidate sample blocks by using the calcification focus segmentation model based on the plurality of candidate sample blocks and corresponding coordinates of the candidate sample blocks in a coordinate system corresponding to the original chest flat-scan CT image. The method adopts a coarse-to-fine strategy, uses a two-stage deep neural network to realize the full-automatic segmentation of the coronary calcifications based on the chest flat scan CT, and can obtain good segmentation results for data with larger noise.

Description

Full-automatic coronary calcified focus segmentation method based on chest flat scan CT

Technical Field

The invention relates to the technical field of medical image segmentation, in particular to a full-automatic segmentation method of coronary calcifications based on chest flat-scan CT.

Background

At present, the morbidity and mortality of cardiovascular diseases are still in a continuously rising state, the mortality is still in the first place, and the old with cardiovascular diseases has reduced immune function, is easy to be infected by other diseases and becomes a critically ill patient, so that early diagnosis of coronary heart disease is particularly necessary. Adequate evidence supports coronary calcification as an independent risk factor for adverse cardiac events in asymptomatic patients.

Quantitative analysis of cardiac Coronary Artery Calcification (CAC) can assess the likelihood of cardiovascular disease. Cardiac CT gated by electrocardio is currently used clinically to assess coronary calcification. The specific post-processing procedure requires the physician to manually select Coronary calcified regions in the image and then the analysis software gives an assessment of the calcification, i.e. Coronary calcification score (CACS), based on the selected regions, which is a semi-automatic procedure. Conventional cardiac calcium scoring scans are often incorporated into coronary CTA examinations, where a complete coronary CTA examination can be several times more expensive than a thoracic CT and the patient is exposed to a much higher dose of radiation. In addition, coronary CTA has certain risks, needs an electrocardiograph gating device and strictly cooperates with respiration of patients, and the coronary CTA examination can not be performed in every level of hospitals. The related art for automatically calculating CACS based on cardiac scout CT or CTA images also has the following drawbacks: a) the existing scheme adopts the image characteristics of the traditional manual design in the characteristic extraction process of the suspected calcified area: histogram of Oriented Gradient (HOG) features, Local Binary Pattern (LBP) features, Haar features, and texture features. b) In the existing scheme, a suspected calcified area is screened through an area threshold, and then a simple classifier is adopted for judging whether the suspected calcified area belongs to calcification or not. The area threshold is set based on certain a priori knowledge, but the a priori knowledge is a rule summarized from most data, and some special cases do not necessarily accord with the rule. If a lesion is just outside the set area threshold, there is no chance of being classified in the existing solution, and therefore a missed detection situation occurs.

The flat scanning CT of the chest is relatively low in cost and wider in coverage rate, and a common physical examination mechanism is also provided with corresponding examination items. If coronary calcification scores could be calculated based on chest scout CT and automatically, the efficiency of clinical diagnosis would be greatly improved (one less cardiac CT, providing additional clinical information to complete cardiopulmonary one-stop screening, and not requiring a physician to manually select calcified regions), while saving the cost of the patient and reducing the radiation dose to which the patient is exposed. However, compared to the above prior art, the calculation of the calcium score based on the chest flat scan CT is more complicated, and especially the chest CT data reconstructed by using the lung window is very noisy. In this case, the classification method (relatively traditional random forest and the like) adopted by the above-mentioned prior art for processing the suspected calcified region can be very challenging on the noisy chest CT.

In view of the above drawbacks of the prior art, there is a need in the art for a fully automatic segmentation scheme of coronary calcifications based on chest flat scan CT.

Disclosure of Invention

In view of this, an embodiment of the present invention provides a method for fully automatically segmenting coronary artery calcification spots based on chest flat-scan CT, which can solve the problem that the prior art cannot achieve full-automatic segmentation of coronary artery calcification spots based on chest flat-scan CT.

Based on the above purpose, an aspect of the embodiments of the present invention provides a full-automatic coronary calcified lesion segmentation method based on chest flat scan CT, including the following steps:

step 1, preprocessing an original chest flat scan CT image to obtain preprocessed data;

step 2, inputting the preprocessed data into a heart segmentation model trained in advance, segmenting a heart region in the original chest flat-scan CT image, and obtaining a heart segmentation image which corresponds to the original chest flat-scan CT image and contains the heart region;

step 3, screening suspected coronary calcifications in the heart area;

step 4, extracting a plurality of candidate sample blocks from the screened suspected coronary calcifications; and

and 5, obtaining a coronary artery calcification focus segmentation result of each candidate sample block by utilizing a pre-trained calcification focus segmentation model based on the candidate sample blocks and corresponding coordinates of the candidate sample blocks in a coordinate system corresponding to the original chest flat-scan CT image.

In some embodiments, the step 1 further comprises:

step 11, zooming the original chest flat scan CT image to a set size; and

and 12, performing normalization operation on the scaled original chest flat-scan CT image.

In some embodiments, the step 3 further comprises:

step 31, restoring the heart segmentation image to the size of the original chest flat scan CT image;

step 32, in the original chest flat scan CT image, determining the boundary of the heart region according to the heart segmentation image with the restored size; and

and step 33, screening suspected coronary calcifications in the boundary of the heart area.

In some embodiments, the step 33 further comprises:

step 331, marking all suspected coronary calcifications in the boundary of the heart region according to the CT value; and

and 332, clustering all the marked suspected coronary calcifications, and screening the suspected coronary calcifications according to the clustering result and corresponding priori knowledge.

In some embodiments, prior to step 5, the method further comprises:

and inputting the candidate sample blocks into the calcification focus segmentation model batch by batch after assembling the candidate sample blocks into batches.

In some embodiments, after the step 5, the method further comprises:

and obtaining a final segmentation result by adopting a voting mechanism based on the coronary artery calcification focus segmentation result of each candidate sample block.

In some embodiments, the training of the cardiac segmentation model comprises the steps of:

collecting chest flat-scan CT images for training and carrying out heart labeling on the images to obtain corresponding heart Mask images; and

and training a first deep neural network by using the chest flat-scan CT image for training and the heart Mask image until a preset first convergence condition is met, so as to obtain the heart segmentation model.

In some embodiments, the training of the calcification focus segmentation model comprises the steps of:

marking coronary artery calcific foci on the chest flat scan CT image for training to obtain a corresponding coronary artery calcific focus Mask image;

extracting a plurality of Mask image sample blocks with a preset size from the coronary calcific lesion Mask image, extracting a plurality of original image sample blocks with the preset size corresponding to the plurality of Mask image sample blocks from the chest flat scan CT image for training, wherein each Mask image sample block and one original image sample block corresponding to each Mask image sample block form a sample pair; and

and training a second deep neural network by using the sample pair until a preset second convergence condition is met, and obtaining the calcification focus segmentation model.

In some embodiments, extracting a plurality of Mask image sample blocks having a preset size from the coronary calcific lesion Mask image comprises the steps of:

clustering the coronary calcific focus Mask images; and

and taking each clustering center as a center, and respectively extracting the Mask image sample blocks.

In some embodiments, extracting a plurality of original image sample blocks having the preset size corresponding to the plurality of Mask image sample blocks from the chest flat scan CT image for training includes the steps of:

and extracting corresponding original image sample blocks from the chest flat-scan CT image for training according to the extracted coordinates of each Mask image sample block.

The invention has the following beneficial technical effects:

the embodiment of the invention provides a two-stage coronary calcified focus full-automatic segmentation method based on chest flat scan CT, which comprises the following steps: in the first stage, firstly, a deep neural network trained under cardiac labeling data is used for segmenting a cardiac region; next, in a second stage, the coronary calcifications in the heart region obtained in the first stage are segmented using another deep neural network trained under coronary calcifications labeling data. The whole process does not need manual participation, the clinical diagnosis efficiency is greatly improved, and more possibilities are provided for popularizing the heart and lung integrated screening. The scheme provided by the invention is suitable for risk assessment of coronary heart disease, is beneficial to improving the accuracy of screening coronary heart disease, improving the prediction efficiency of individual adverse cardiovascular events of patients, assisting clinical formulation of an individualized comprehensive treatment scheme and reducing the risk of adverse cardiovascular events of patients.

The invention adopts the current successful deep neural network to automatically extract the image characteristics to replace the traditional manual characteristics, which is an obvious trend in the current image processing task.

After a heart area is obtained in the first stage, clustering is carried out on suspected calcified areas, then preliminary screening is carried out on the suspected calcified areas by combining an area threshold, sample blocks are extracted by taking a clustering center as the center after preliminary screening (due to the fact that the space sizes of the sample blocks are relatively large, the sample blocks are probably overlapped), then the sample blocks are predicted by a trained deep segmentation network, calcific foci which are excluded because the area threshold are still probably found back in the neural network reasoning process, and the prediction results of the sample blocks with certain spatial overlapping finally determine whether each point in the space is a real calcified point or not through a voting mechanism. This allows more robust results to be achieved than with a simple classifier. In summary, 1) the depth segmentation network of the second stage of the present invention can further narrow the setting of the area threshold; 2) the voting mechanism makes the segmentation result more robust.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other embodiments can be obtained by using the drawings without creative efforts.

FIG. 1 is a schematic flow chart of a fully automatic coronary calcification focus segmentation method based on chest flat scan CT according to an embodiment of the present invention;

FIG. 2 is a schematic flow diagram of a pre-processing procedure of an original chest flat scan CT image;

FIG. 3 is a schematic flow diagram of a process for screening suspected coronary calcifications in a region of the heart;

FIG. 4 is a schematic flow diagram of screening for suspected coronary calcifications within the borders of a heart region;

FIG. 5 is a schematic flow chart diagram of a training process for a heart segmentation model;

FIG. 6 is a schematic diagram of annotation data of a heart segmentation model;

FIG. 7 is a schematic flow chart of a training process of a calcium focus segmentation model;

FIG. 8 is a schematic diagram of labeled data of a calcification focus segmentation model; and

fig. 9 is a comparison of the results achieved when the method of the present invention is used to perform an actual segmentation task.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the following embodiments of the present invention are described in further detail with reference to the accompanying drawings.

It should be noted that all expressions of "first", "second", and the like in the embodiments of the present invention are used for distinguishing a plurality of entities with the same name but different names or different parameters, and it is understood that "first", "second", and the like are only for convenience of description and should not be construed as limiting the embodiments of the present invention, and the descriptions in the following embodiments are omitted.

Based on the above purpose, the present invention provides an embodiment of a full-automatic coronary calcification focus segmentation method based on chest flat scan CT. Fig. 1 shows a schematic flow diagram of the method. The method is mainly based on a segmentation model commonly used in deep learning, such as UNet, however, it is obvious to those skilled in the art that the segmentation network used in the present invention may be other segmentation networks capable of achieving the purpose of the present invention besides the basic UNet. The present invention implements the coronary calcification spot segmentation scheme in two main stages, because the calcification spots of the heart coronary belong to very tiny targets in the whole chest flat-scan CT, and the proportion of pixels occupied by the calcification spots is about one ten thousandth less in a 512X 100 three-dimensional CT image. Therefore, to accomplish such a segmentation task, the present invention adopts the coarse-to-fine strategy. In the first stage, the method obtains a heart region through a heart segmentation model, then based on the obtained heart region, in the second stage, a more refined segmentation operation is carried out through a calcification focus segmentation model, and finally, a final coronary artery calcification focus segmentation result is obtained.

Specifically, as shown in fig. 1, the fully automatic coronary calcification focus segmentation method based on chest flat scan CT includes the following steps:

and step S1, preprocessing the original chest flat scan CT image to obtain preprocessed data.

When coronary artery calcification focus segmentation operation needs to be executed, the early-stage chest flat scan CT information is processed to obtain DICOM (Digital Imaging and Communications in Medicine) format data, each DICOM image data comprises image data of a slice of a certain layer of the chest, and for convenience of storage and management, the multi-layer DICOM image data of a patient can be merged into nii Digital format files to store the original chest flat scan CT image. The original chest flat scan CT image in the nii file is then pre-processed.

Step S2, inputting the preprocessed data into a pre-trained heart segmentation model, and segmenting a heart region in the original chest flat-scan CT image to obtain a heart segmentation image including the heart region corresponding to the original chest flat-scan CT image.

This step is a first stage segmentation, i.e. a heart segmentation, which may also be understood as a coarse segmentation. To perform the segmentation of the heart region, a suitable deep neural network (e.g., UNet) is first selected to construct a heart segmentation model, and the training process of the heart segmentation model will be described in detail below.

And step S3, screening suspected coronary calcifications in the heart area.

Before the segmentation operation of the second stage is executed, all suspected coronary calcifications in the heart region need to be preliminarily screened to filter out a part of suspected calcifications, so that the burden of the segmentation model of the second stage is reduced.

And step S4, extracting a plurality of candidate sample blocks from the screened suspected coronary calcifications.

And (4) respectively extracting corresponding candidate sample blocks by taking each clustering center as a center for the screened suspected coronary calcification foci. The candidate sample block has a size of (16, 64, 64) (i.e., the input size of the second stage segmentation model), but the size is not necessarily (16, 64, 64), and those skilled in the art can make a matching setting according to the GPU configuration and the segmentation result.

And step S5, obtaining a coronary artery calcification focus segmentation result of each candidate sample block by utilizing a pre-trained calcification focus segmentation model based on the candidate sample blocks and corresponding coordinates thereof in a coordinate system corresponding to the original chest flat-scan CT image.

This step is a second stage of segmentation, namely coronary calcific lesion segmentation, which can also be understood as fine segmentation. To realize the segmentation of coronary artery calcifications, a suitable deep neural network (e.g., UNet) needs to be selected to construct a calcifications segmentation model, and the training process of the calcifications segmentation model will be described in detail below.

Fig. 2 shows a schematic flow diagram of the preprocessing procedure of an original chest flat scan CT image. As shown in fig. 2, the image preprocessing process of step 1 further includes:

step S11, scaling the original chest flat scan CT image to a set size. Since the acquired CT images have heterogeneity due to the inconsistency of parameters of different CT data acquisition devices, nii data need to be transformed into a uniform size (which can be realized by Reshape function) by interpolation operation, for example, the size may be (64, 256, 256) (i.e., the input size of the first stage segmentation model), but the size is not necessarily (64, 256, 256), and those skilled in the art can make matching settings according to the GPU configuration and the segmentation result.

Step S12, a normalization operation is performed on the scaled original chest flat scan CT image, i.e. the data is transformed to [0, 1] space, and then can be input into the trained heart segmentation model.

Fig. 3 shows a schematic flow chart of a process of screening a suspected coronary calcification region in a heart region. As shown in fig. 3, the step S3 further includes:

and step S31, restoring the heart segmentation image to the size of the original chest flat scan CT image. In this step, the heart segmentation image inferred by the heart segmentation model needs to be restored to the original size again through an interpolation operation (which can be realized again through the Reshape function).

Step S32, in the original chest flat-scan CT image, determining the boundary of the heart region according to the heart segmentation image after size recovery, which can be represented by a rectangular bounding box, and then recording the coordinates of one corner point of the bounding box relative to the entire CT.

And step S33, screening suspected coronary calcifications in the boundary of the heart area. Specifically, as shown in fig. 4, this step further includes: step S331, marking all suspected coronary calcifications in the boundary of the heart region according to the CT value, specifically, setting a threshold (for example, 130HU (hounsfield unit)), and then, for each pixel point in the heart region, if the CT value is greater than 130HU, marking as a suspected calcifications, otherwise, marking as a background; and S332, clustering all the marked suspected coronary calcifications, and screening the suspected coronary calcifications according to a clustering result and corresponding priori knowledge, wherein each class in the clustering result has a clustering center and a corresponding area, and some suspected calcifications can be preliminarily filtered according to the corresponding priori knowledge (for example, experience tells that the coronary calcifications are in a certain range).

Next, in step S4, for the suspected coronary calcifications left after the preliminary screening, a plurality of candidate sample blocks can be extracted with the cluster center as the center, and their coordinates in the heart BoundingBox are converted into the original CT coordinate system and recorded.

When there are many extracted candidate sample blocks, it is inefficient if inference is performed by serially inputting a single sample block (Patch) into the calcification focus segmentation model. Therefore, in order to accelerate the reasoning process of the calcification focus segmentation model, the candidate sample blocks can be assembled into a plurality of batches (each Batch (Batch) can contain 128 samples or even more depending on the configuration of the GPU), and then the candidate sample blocks are input into the calcification focus segmentation model Batch by Batch for parallel reasoning, so that the purpose of accelerating the reasoning speed is achieved. And after the segmentation result of each candidate sample block is obtained through the reasoning of the calcification focus segmentation model, the segmentation results are assembled back to the original CT coordinate system one by one, and then the coronary calcification focus segmentation result based on the chest flat-scan CT image can be obtained.

Since there may be overlap between different candidate sample blocks, in order to make the result more robust, the final calcification segmentation result may be decided according to a Voting mechanism, and a Majority Voting algorithm (Majority Voting) may be used here.

Fig. 5 shows a schematic flow chart of a training process of a heart segmentation model. As shown in fig. 5, the training of the heart segmentation model comprises the following steps: step S51, collecting chest flat scan CT images for training and carrying out heart labeling on the images to obtain corresponding heart Mask images; and step S52, training a first deep neural network by using the chest flat scan CT image for training and the cardiac Mask image until a preset first convergence condition is met, and obtaining the cardiac segmentation model.

An example of training a heart segmentation model is illustrated below. First, hundreds of chest scans of CT data are collected from a hospital, each containing varying degrees of coronary calcification. Then, using these data, hundreds of cases of cardiac masks are labeled for training a deep neural network cardiac segmentation model, i.e., in each chest flat scan CT, the pixels of the cardiac region are labeled 1, the aorta is labeled 2, and the background is labeled 0, as shown in fig. 6. Each chest scan CT data has a corresponding labeled cardiac Mask image. Training of the specified model can then be performed. The training process is that a deep neural network is given, network parameters are initialized firstly, then the network carries out forward operation on given input data to obtain an output result, the difference between the output result and labeled data is calculated, and then the difference is reversely propagated back to the network to carry out parameter updating so as to achieve the aim of network training. After the network parameters converge, the expected output result can have a tangible form of expression with the annotation data for a given input. The difference calculation in the training process is generally calculated through a loss function, the loss function used by the method is TverskyLoss, which is generalized DiceLoss, the superposition degree can be directly calculated aiming at the network output result and the labeled Mask, and the form is as follows:

the formula expresses a form of two-class segmentation, and can be expanded into a plurality of forms when being specifically realized, for example, the heart segmentation model of the invention is aimed at the case of three-class segmentation.

It should be noted that for the breast CT data with large size heterogeneity, all data need to be reshape to a uniform size, for example (64, 256, 256), before being input to the network for training. Of course, the data pre-processing stage also includes the common random cropping, random mirroring, and normalization operations. The 200 rounds were then trained using an Adam optimizer with initial learning rate set to 2e-4, momentum parameter β₁Is set to 0.5, beta₂Set to 0.99. The BatchSize (batch size) is set to 4.

Fig. 7 shows a schematic flow chart of a training process of a calcium focus segmentation model. As shown in fig. 7, the training of the calcification segmentation model includes the following steps: step S71, coronary artery calcification focus labeling is carried out on the chest flat-scan CT image for training to obtain a coronary artery calcification focus Mask image corresponding to the chest flat-scan CT image; step S72, extracting a plurality of Mask image sample blocks with preset sizes from the coronary calcific lesion Mask image, extracting a plurality of original image sample blocks with the preset sizes corresponding to the plurality of Mask image sample blocks from the chest flat-scan CT image for training, wherein each Mask image sample block and one original image sample block corresponding to the Mask image sample block form a sample pair; and step S73, training a second deep neural network by using the sample pair until a preset second convergence condition is met, and obtaining the calcification focus segmentation model. Wherein, extracting a plurality of Mask image sample blocks with preset sizes from the coronary calcific lesion Mask image specifically comprises: clustering the coronary calcific focus Mask images; and taking each cluster center as a center, and respectively extracting the Mask image sample blocks. Extracting a plurality of original image sample blocks with the preset size corresponding to the plurality of Mask image sample blocks from the chest flat scan CT image for training specifically comprises: and extracting corresponding original image sample blocks from the chest flat-scan CT image for training according to the extracted coordinates of each Mask image sample block.

Here, the following strategy is adopted for the collection of training data of the calcium lesion segmentation model. First, hundreds of cases of coronary calcific foci Mask were labeled for training another deep neural network model, as shown in fig. 8. Then, coronary calcific foci Mask is clustered, sample blocks with preset sizes (16, 64 and 64) are extracted by taking each clustering center as a center, and the extracted reference coordinates are synchronously implemented into the original image, so that N sample pairs with preset sizes are formed (one original image sample block and the corresponding Mask sample block form one sample pair). And finally training a calcification focus segmentation model based on the sample pairs. The specific training process is similar to the above-illustrated training process of the heart segmentation model, and is not described herein again.

Preferably, after the neural network reasoning is completed, the present invention also sets some post-processing operations, mainly aiming at eliminating false positives. The false positives can be classified into two categories, one is false positive in the rib region (the HU value in the bone region is generally high), and the other is false positive on the aorta (this is why the above embodiment labels the aorta alone when labeling the cardiac Mask). In order to achieve the purpose, the operation of setting the non-1 to 0 is carried out on the segmented heart Mask, and the two false positives can be removed by carrying out bit multiplication on the segmented heart Mask and the calcification segmentation result.

The invention can realize full-automatic segmentation of coronary artery calcification focuses based on chest flat scan CT by using a two-stage deep segmentation network, and has good segmentation results for data with larger noise. The heart segmentation model of the first stage considers the aorta to be segmented into a category separately, and provides a basis for removing false positive on the aorta for the result of the calcification segmentation of the second stage. In addition, the small target segmentation of calcification is an innovation in the task of calcification segmentation by adopting a mode of clustering first and then extracting training samples. Fig. 9 shows the effect achieved when the method of the invention is used to perform the actual segmentation task. It can be seen from the comparison of the original coronary calcifications image, the Ground Truth and the segmentation result shown in fig. 9 that the expected segmentation effect is achieved by the present invention.

It should be noted that, as will be understood by those skilled in the art, all or part of the processes in the operations of the above embodiments may be implemented by a computer program, which may be stored in a computer-readable storage medium, and when executed, may include the processes of the above embodiments. The computer program may achieve the same or similar effects as the corresponding foregoing operational embodiments.

Further, it should be understood that the computer-readable storage medium (e.g., memory) employed to implement the operations of the present invention may be either volatile memory or nonvolatile memory, or may include both volatile and nonvolatile memory. By way of example, and not limitation, nonvolatile memory can include Read Only Memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM), which can act as external cache memory. By way of example and not limitation, RAM is available in a variety of forms such as synchronous RAM (DRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), and Direct Rambus RAM (DRRAM). The storage devices of the disclosed aspects are intended to comprise, without being limited to, these and other suitable types of memory.

Those of skill would further appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the disclosure herein may be implemented or performed with the following components which are designed to perform the functions described herein: a general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination of these components. A general purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. A processor may also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP, and/or any other such configuration.

In one or more exemplary designs, the functions may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A storage media may be any available media that can be accessed by a general purpose or special purpose computer. By way of example, and not limitation, such computer-readable media can comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a general-purpose or special-purpose computer, or a general-purpose or special-purpose processor. Also, any connection is properly termed a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, Digital Subscriber Line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of medium. Disk and disc, as used herein, includes Compact Disc (CD), laser disc, optical disc, Digital Versatile Disc (DVD), floppy disk, blu-ray disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

The above is an exemplary embodiment of the present disclosure, and the order of disclosure of the above embodiment of the present disclosure is only for description and does not represent the merits of the embodiment. It should be noted that the discussion of any embodiment above is exemplary only, and is not intended to intimate that the scope of the disclosure, including the claims, of embodiments of the invention is limited to those examples, and that various changes and modifications may be made without departing from the scope, as defined in the claims. The functions, steps and/or actions of the claims in accordance with the disclosed embodiments described herein need not be performed in any particular order. Furthermore, although elements of the disclosed embodiments of the invention may be described or claimed in the singular, the plural is contemplated unless limitation to the singular is explicitly stated.

Claims

1. A full-automatic segmentation method of coronary calcified foci based on chest flat scan CT is characterized by comprising the following steps:

step 3, screening suspected coronary calcifications in the heart area;

step 5, obtaining coronary artery calcification focus segmentation results of each candidate sample block by utilizing a pre-trained calcification focus segmentation model based on the candidate sample blocks and corresponding coordinates thereof in a coordinate system corresponding to the original chest flat-scan CT image,

wherein the training of the heart segmentation model comprises the steps of:

training a first deep neural network by using the chest flat-scan CT image for training and the cardiac Mask image until a preset first convergence condition is met to obtain the cardiac segmentation model,

the training of the calcification focus segmentation model comprises the following steps:

2. The method for fully automatically segmenting coronary calcifications based on chest flat scan CT as claimed in claim 1, wherein said step 1 further comprises:

step 11, zooming the original chest flat scan CT image to a set size; and

3. The method for fully automatically segmenting coronary calcifications based on chest flat scan CT as claimed in claim 2, wherein said step 3 further comprises:

4. The method for fully automatically segmenting coronary calcifications based on thoracic flat scan CT as claimed in claim 3, wherein said step 33 further comprises:

5. The method for fully automatically segmenting coronary calcifications based on thoracic flat scan CT as claimed in claim 4, wherein before the step 5, the method further comprises:

6. The fully automatic coronary calcification lesion segmentation method based on thoracic flat scan CT as claimed in claim 1, wherein after the step 5, the method further comprises:

7. The coronary artery calcification focus full-automatic segmentation method based on the chest flat scan CT as claimed in claim 1, wherein the step of extracting a plurality of Mask image sample blocks with preset sizes from the coronary artery calcification focus Mask image comprises the following steps:

clustering the coronary calcific focus Mask images; and

8. The method for fully automatically segmenting coronary calcifications based on thoracic flat scan CT as claimed in claim 1, wherein the step of extracting a plurality of original image sample blocks with the preset size corresponding to the plurality of Mask image sample blocks from the chest flat scan CT image for training comprises the steps of: