CN117649418A - Chest multi-organ segmentation method and system and computer readable storage medium - Google Patents
Chest multi-organ segmentation method and system and computer readable storage medium Download PDFInfo
- Publication number
- CN117649418A CN117649418A CN202410128271.9A CN202410128271A CN117649418A CN 117649418 A CN117649418 A CN 117649418A CN 202410128271 A CN202410128271 A CN 202410128271A CN 117649418 A CN117649418 A CN 117649418A
- Authority
- CN
- China
- Prior art keywords
- chest
- organ
- processing
- segmentation
- marked
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 230000011218 segmentation Effects 0.000 title claims abstract description 133
- 238000000034 method Methods 0.000 title claims abstract description 56
- 210000000056 organ Anatomy 0.000 claims abstract description 158
- 238000012545 processing Methods 0.000 claims abstract description 141
- 238000012549 training Methods 0.000 claims abstract description 32
- 210000000038 chest Anatomy 0.000 claims description 115
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 42
- 210000000115 thoracic cavity Anatomy 0.000 claims description 40
- 230000008569 process Effects 0.000 claims description 31
- 238000004590 computer program Methods 0.000 claims description 14
- 238000007781 pre-processing Methods 0.000 claims description 10
- 238000012952 Resampling Methods 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 7
- 238000005070 sampling Methods 0.000 claims description 6
- 239000003550 marker Substances 0.000 claims description 5
- 238000010276 construction Methods 0.000 abstract description 2
- 210000004072 lung Anatomy 0.000 description 9
- 210000000278 spinal cord Anatomy 0.000 description 7
- 210000002216 heart Anatomy 0.000 description 5
- 206010028980 Neoplasm Diseases 0.000 description 4
- 210000003238 esophagus Anatomy 0.000 description 4
- 210000003437 trachea Anatomy 0.000 description 4
- 230000009466 transformation Effects 0.000 description 4
- 210000000481 breast Anatomy 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 238000002591 computed tomography Methods 0.000 description 3
- 238000001959 radiotherapy Methods 0.000 description 3
- 230000004927 fusion Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000011282 treatment Methods 0.000 description 2
- 241000699670 Mus sp. Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 201000011510 cancer Diseases 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 210000000920 organ at risk Anatomy 0.000 description 1
- 238000011470 radical surgery Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000004083 survival effect Effects 0.000 description 1
- 238000002759 z-score normalization Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The invention relates to the technical field of medical image processing, in particular to a chest multi-organ segmentation method and system and a computer readable storage medium; according to the invention, the STUNet model is adopted as a basic model of the segmentation model, the chest CT image marked by partial organs is trained, the pseudo label is given to the missing mark data, and finally, the non-marked or missing mark chest CT image is formed into a full-marked CT image, and then the chest multi-organ segmentation model is obtained through training, so that the segmentation of the non-marked or missing mark chest CT image is realized, the problem of less full data of the chest CT image is solved, the missing mark or non-marked chest CT image is fully utilized, and the accurate construction of the chest multi-organ segmentation model is realized.
Description
Technical Field
The invention relates to the technical field of medical image processing, in particular to a chest multi-organ segmentation method and system and a computer readable storage medium.
Background
In the medical field, accurate radiotherapy technology has greatly improved survival rates of cancer patients. However, these advanced treatments require not only accurate determination of the contours of the target tumor, but also accurate identification of the contours of the vital organs surrounding the tumor, which allows these organs to be protected during radiation therapy. In addition, in the field of surgical application, accurate pre-operation assessment and normative radical surgery are important measures for improving the curative effect of tumor diagnosis and treatment, and organ contour identification based on CT image data can help doctors to rapidly, accurately and consistently finish the step of organ sketching in operation planning.
The identification of the radiotherapy organs at risk in the current clinical CT images is mainly obtained by manual sketching by doctors. Manual delineation by a physician thus has the following drawbacks: 1. the sketching efficiency is low; 2. severely relying on the clinical experience of doctors; 3. the repeatability is poor, and the sketched results of different doctors in different states at different times are inconsistent.
At present, chest ct has certain image data, but marking data is lacking, marking organs are not uniform, some marked lungs and spinal cords, some marked hearts and spinal cords are marked as needed, and how to use the data of lack of marking, omission of marking and no marking to improve the performance of a model to a certain extent is a great challenge.
Disclosure of Invention
In view of the foregoing deficiencies of the prior art, the present invention is directed to a method and system for segmentation of multiple organs in the breast and a computer readable storage medium for achieving accurate segmentation of multiple organs in the breast using CT images of non-labeled, unlabeled and unlabeled breast organs.
In order to solve the problems, the invention adopts the following technical scheme:
a chest multi-organ segmentation method, comprising:
collecting CT images of the chest organ marks, dividing the CT images of the same marked organ into the same group of data sets, and sorting according to the number of the marked organs to obtain CT image data sets of a plurality of chest organ marks;
training a first segmentation model by adopting a CT image data set with the most thoracic organ marks, predicting a data set with the number of marked organs being less than that of the CT image data set with the most thoracic organ marks by adopting the trained first segmentation model, and endowing the data set with pseudo tags of the uniquely marked organs of the CT image data set with the most thoracic organ marks by adopting a prediction result;
the CT image dataset with the most thoracic organ marks and the CT image dataset with the second most thoracic organ marks are fused to train a segmentation model II, then the trained segmentation model II is adopted to predict the datasets with the number of marked organs being less than that of the CT image dataset with the second most thoracic organ marks, and the predicting result is adopted to endow the datasets with pseudo labels of the CT image dataset with the most thoracic organ marks and the CT image dataset with the second most thoracic organ marks together and uniquely mark the organs;
continuing to fuse the data sets to train the segmentation model to obtain different segmentation models and CT image data sets of a plurality of chest organ full markers corresponding to the CT image data sets of a plurality of chest organ markers;
applying pseudo labels to chest CT images of the newly incorporated organs without full markers by adopting the different segmentation models, and combining the pseudo labels with CT image data sets of the full markers of the plurality of chest organs to form a CT image data set of the full markers;
training a segmentation model by adopting the fully-marked CT image data set to obtain a chest multi-organ segmentation model;
and segmenting the chest CT image data of the organs which are not fully marked by adopting the chest multi-organ segmentation model.
As an embodiment, the preprocessing is performed after the CT image of the thoracic organ marker is acquired; wherein the preprocessing includes image cropping, resampling, normalization and data enhancement.
As an implementation manner, the segmentation model adopts a STUNet model as a basic model;
the STUNet model sequentially comprises a stem block, 13 residual blocks and a seg head block to form a U-shaped structure, wherein a downsampling process is carried out among the first 6 residual blocks to form an encoder, an upsampling process is carried out among the last 6 residual blocks to form a decoder, and the residual blocks in the encoder are correspondingly connected with the residual blocks in the decoder in an end-to-end manner;
the step block comprises two processing branches, wherein one processing branch is convolution processing of 1 multiplied by 1, the other processing branch is convolution processing of two pieces of 3 multiplied by 3, and the two processing branches are combined and then used as output of the step block;
the residual block comprises two processing branches, wherein one processing branch is subjected to convolution processing of two 3 multiplied by 3 and is combined with the other processing branch to be used as output of the residual block;
the seg head block is output after being subjected to convolution processing of 1 multiplied by 1;
the downsampling process comprises two processing branches, wherein one processing branch is convolution processing with the step length of 2 being 1 multiplied by 1, and the other processing branch is convolution processing with the step length of 2 being 3 multiplied by 3 and convolution processing with the step length of 2 being 3 multiplied by 3, and the two processing branches are combined and then used as output of the downsampling process;
the up-sampling process includes nearest neighbor interpolation processing and 1×1 with step size of 1 convolution processing of x 1.
As an embodiment, the different segmentation models have a depth of (2, 2) and a width of (64, 128, 256, 512, 1024, 1024); the chest multi-organ segmentation model has a depth of (1, 1) and a width of (32, 64, 128, 256, 512, 512).
A chest multi-organ segmentation system comprises a data acquisition processing module, a model training and pseudo-label giving module, a full-label data set processing module, a chest multi-organ segmentation model training module and a chest multi-organ segmentation module;
the data acquisition processing module is used for acquiring CT images of the chest organ marks, dividing the CT images of the same marked organ into the same group of data sets, and sorting the CT images according to the number of the marked organs to obtain CT image data sets of a plurality of chest organ marks;
the model training and pseudo-label giving module is used for training a segmentation model I by adopting a CT image data set with the most chest organ marks, predicting a data set with the number of marked organs being less than that of the CT image data set with the most chest organ marks by adopting the trained segmentation model I, and giving a pseudo-label of a uniquely marked organ to the CT image data set with the most chest organ marks by adopting a prediction result;
the CT image dataset with the most thoracic organ marks and the CT image dataset with the second most thoracic organ marks are fused to train a segmentation model II, then the trained segmentation model II is adopted to predict the datasets with the number of marked organs being less than that of the CT image dataset with the second most thoracic organ marks, and the predicting result is adopted to endow the datasets with pseudo labels of the CT image dataset with the most thoracic organ marks and the CT image dataset with the second most thoracic organ marks together and uniquely mark the organs;
continuing to fuse the data sets to train the segmentation model to obtain different segmentation models and CT image data sets of a plurality of chest organ full markers corresponding to the CT image data sets of a plurality of chest organ markers;
the full-label data set processing module is used for endowing pseudo labels to chest CT images of the newly-incorporated organs without full labels by adopting the different segmentation models, and combining the pseudo labels with CT image data sets of the full labels of the chest organs to form a full-label CT image data set;
the chest multi-organ segmentation model training module is used for training a segmentation model by adopting the fully-marked CT image data set to obtain a chest multi-organ segmentation model;
the chest multi-organ segmentation module is used for segmenting chest CT image data of the organs which are not fully marked by adopting the chest multi-organ segmentation model.
As an implementation manner, the data acquisition processing module acquires CT images of the chest organ markers and then performs preprocessing; wherein the preprocessing includes image cropping, resampling, normalization and data enhancement.
As an implementation manner, the segmentation model adopts a STUNet model as a basic model;
the STUNet model sequentially comprises a stem block, 13 residual blocks and a seg head block to form a U-shaped structure, wherein a downsampling process is carried out among the first 6 residual blocks to form an encoder, an upsampling process is carried out among the last 6 residual blocks to form a decoder, and the residual blocks in the encoder are correspondingly connected with the residual blocks in the decoder in an end-to-end manner;
the step block comprises two processing branches, wherein one processing branch is convolution processing of 1 multiplied by 1, the other processing branch is convolution processing of two pieces of 3 multiplied by 3, and the two processing branches are combined and then used as output of the step block;
the residual block comprises two processing branches, wherein one processing branch is subjected to convolution processing of two 3 multiplied by 3 and is combined with the other processing branch to be used as output of the residual block;
the seg head block is output after being subjected to convolution processing of 1 multiplied by 1;
the downsampling process comprises two processing branches, wherein one processing branch is convolution processing with the step length of 2 being 1 multiplied by 1, and the other processing branch is convolution processing with the step length of 2 being 3 multiplied by 3 and convolution processing with the step length of 2 being 3 multiplied by 3, and the two processing branches are combined and then used as output of the downsampling process;
the up-sampling process includes nearest neighbor interpolation processing and 1×1 with step size of 1 convolution processing of x 1.
As an embodiment, the different segmentation models have a depth of (2, 2) and a width of (64, 128, 256, 512, 1024, 1024); the chest multi-organ segmentation model has a depth of (1, 1) and a width of (32, 64, 128, 256, 512, 512).
An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the method when the computer program is executed.
A computer readable storage medium storing a computer program which, when executed by a processor, implements the steps of the method.
The invention has the beneficial effects that: according to the invention, a STUNet model is adopted as a basic model of a segmentation model, a chest CT image marked by partial organs is trained, pseudo labels are given to the missing mark data, and finally, the non-marked or missing mark chest CT image is formed into a full-marked CT image, and then, the chest multi-organ segmentation model is obtained through training, the segmentation of the non-marked or missing mark chest CT image is realized, the problem of less full-marked data of the chest CT image is solved, the missing mark or non-marked chest CT image is fully utilized, and the accurate construction of the chest multi-organ segmentation model is realized.
Drawings
Fig. 1 is a flowchart of a chest multi-organ segmentation method according to an embodiment of the present invention.
Fig. 2 is a schematic diagram of a chest multi-organ segmentation system according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to specific examples.
It should be noted that these examples are only for illustrating the present invention, and not for limiting the present invention, and simple modifications of the method under the premise of the inventive concept are all within the scope of the claimed invention.
Referring to fig. 1, a chest multi-organ segmentation method is provided, comprising:
s100, acquiring CT images of the chest organ marks, dividing the CT images of the same marked organs into the same group of data sets, and sorting according to the number of the marked organs to obtain CT image data sets of a plurality of chest organ marks.
Such as using existing data and published data sets to form the following data sets:
data set one: the markers comprise six organs including left lung, right lung, heart, spinal cord, trachea and esophagus;
data set two: the markers include five organs of left lung, right lung, heart, spinal cord, trachea;
data set three: the markers include four organs of left lung, right lung, heart, spinal cord;
etc.
The CT image of the thoracic organ marker is acquired and then preprocessed, wherein the preprocessing comprises image cropping, resampling, normalization and data enhancement.
Image clipping: the training CT scan is cropped along the z-axis based on the group trunk or pseudo tag. Specifically, first, the index of the start slice and the end slice including the target region is calculated based on the label. To preserve the context information of the split targets, we decrease the index of the start tile by 10 and increase the index of the end tile by 10.
Anisotropic data resampling: image redirection is performed to the desired direction and then all CT scans are resampled to match the median voxel spacing of the training dataset. The image resampling adopts third-order spline interpolation, and the label resampling adopts nearest neighbor interpolation.
CT image normalization: the pixel values are collected in a cropped CT scan and then all data is truncated to fall within the [0.5,99.5] range of foreground voxel values. Then, z-score normalization was applied.
Data enhancement, namely adopting methods of random rotation, random scaling, elastic transformation, brightness transformation, contrast transformation, gamma transformation and the like to enhance the data.
S200, training a segmentation model I by using the CT image data set with the most thoracic organ marks, for example, training the segmentation model I by using the data set I.
And predicting a data set (a data set II, a data set III and the like) with the number of marked organs being less than that of the CT image data set with the most marked organs of the chest organs by adopting a trained segmentation model, and endowing the data set (the data set II, the data set III and the like) with a pseudo label of the CT image data set with the most marked organs of the chest organs by adopting a prediction result. A uniquely labeled organ of the dataset is then the "esophagus".
And (3) merging the CT image dataset with the most chest organ marks (namely, the dataset I) and the CT image dataset with the second most chest organ marks (namely, the dataset II), training a segmentation model II (namely, the dataset I and the dataset II correspond to the segmentation model II), predicting the dataset (and the dataset III, etc.) with the number of marked organs smaller than that of the CT image dataset with the second most chest organ marks by adopting the trained segmentation model II, and endowing the dataset with pseudo tags of the marked organs together and exclusively by adopting the prediction result. The organ labeled by both is the "trachea".
Based on the thought of the fusion data set and the training model, continuing to train the segmentation model by the fusion data set to obtain different segmentation models and CT image data sets of a plurality of chest organ full markers corresponding to the CT image data sets of a plurality of chest organ markers. Through the above flow, the data of the missing marks in the first data set, the second data set, the third data set and the like are all given with pseudo labels, and different segmentation models are obtained, for example, the segmentation models are used for segmenting esophagus, the segmentation models are used for segmenting trachea, and the like, and the subsequently constructed segmentation models are used for segmenting spinal cord and the like. By the method, training data required by training the segmentation model can be reduced, and the segmentation accuracy of the segmentation model can be ensured when the training data is small.
And S300, adopting the different segmentation models to assign pseudo labels to the chest CT images of the newly incorporated organs which are not fully marked, and combining the pseudo labels with the CT image data sets of the chest organs which are fully marked to form a CT image data set of the full mark.
The invention relates to a method for dividing CT image data of a chest organ, which comprises the steps of dividing the chest organ into different divided models and dividing each model into a plurality of parts, wherein the CT image data of the chest organ has more marked data and more unmarked data, and the different divided models are generally used for dividing the chest organ, and the part of the CT image data is marked with pseudo labels, so that all CT images are fully marked CT image data.
S400, training a segmentation model by using the full-labeled CT image data set to obtain a chest multi-organ segmentation model.
The segmentation model one, two and chest multi-organ segmentation model all adopt a STUNet model as a basic model.
However, the different segmentation models "segmentation models one, two" have a depth of (2, 2) and a width of (64, 128, 256, 512, 1024, 1024); the chest multi-organ segmentation model has a depth of (1, 1) and a width of (32, 64, 128, 256, 512, 512). The method aims at scaling the finally obtained chest multi-organ segmentation model so as to reduce the memory requirement and accelerate the reasoning speed.
The STUNet model sequentially comprises a step block, 13 residual blocks and a seg head block to form a U-shaped structure, wherein a downsampling process is carried out among the first 6 residual blocks to form an encoder, an upsampling process is carried out among the last 6 residual blocks to form a decoder, and the residual blocks in the encoder and the residual blocks in the decoder are correspondingly connected end to end;
the step block comprises two processing branches, wherein one processing branch is convolution processing of 1 multiplied by 1, the other processing branch is convolution processing of two pieces of 3 multiplied by 3, and the two processing branches are combined and then used as output of the step block;
the residual block comprises two processing branches, wherein one processing branch is subjected to convolution processing of two 3 multiplied by 3 and is combined with the other processing branch to be used as output of the residual block;
the seg head block is output after being subjected to convolution processing of 1 multiplied by 1;
the downsampling process comprises two processing branches, wherein one processing branch is convolution processing with the step length of 2 being 1 multiplied by 1, and the other processing branch is convolution processing with the step length of 2 being 3 multiplied by 3 and convolution processing with the step length of 2 being 3 multiplied by 3, and the two processing branches are combined and then used as output of the downsampling process;
the up-sampling process includes nearest neighbor interpolation processing and 1×1 with step size of 1 convolution processing of x 1.
S500, segmenting chest CT image data of the organs which are not fully marked by adopting the chest multi-organ segmentation model.
Based on the above method, the individual organ segmentation capability of the finally constructed chest multi-organ segmentation model is shown in table 1.
TABLE 1 organ segmentation Capacity Table
| Organ | Spinal cord | Right lung | Left lung | Heart and method for producing the same | Esophagus | Air pipe | Average of |
| Dice | 0.816 | 0.975 | 0.974 | 0.916 | 0.794 | 0.840 | 0.886 |
Referring to fig. 2, a chest multi-organ segmentation system is provided, which includes a data acquisition processing module 100, a model training and pseudo-tag assignment module 200, a full-label data set processing module 300, a chest multi-organ segmentation model training module 400, and a chest multi-organ segmentation module 500;
the data acquisition processing module 100 is configured to acquire CT images of the thoracic organ markers, divide the CT images of the same marker organs into the same group of data sets, and sort the data sets according to the number of the marker organs to obtain CT image data sets of the plurality of thoracic organ markers;
the model training and pseudo-label giving module 200 is configured to train a segmentation model one by using a CT image dataset with the most thoracic organ markers, predict a dataset with the number of marked organs less than that of the CT image dataset with the most thoracic organ markers by using the trained segmentation model one, and give the dataset pseudo-labels of the uniquely marked organs of the CT image dataset with the most thoracic organ markers by using the prediction result;
the CT image dataset with the most thoracic organ marks and the CT image dataset with the second most thoracic organ marks are fused to train a segmentation model II, then the trained segmentation model II is adopted to predict the datasets with the number of marked organs being less than that of the CT image dataset with the second most thoracic organ marks, and the predicting result is adopted to endow the datasets with pseudo labels of the CT image dataset with the most thoracic organ marks and the CT image dataset with the second most thoracic organ marks together and uniquely mark the organs;
continuing to fuse the data sets to train the segmentation model to obtain different segmentation models and CT image data sets of a plurality of chest organ full markers corresponding to the CT image data sets of a plurality of chest organ markers;
the full-labeled data set processing module 300 is configured to assign pseudo labels to the chest CT images of the newly incorporated non-full-labeled organs by using the different segmentation models, and combine the pseudo labels with the CT image data sets of the full-labeled chest organs to form a full-labeled CT image data set;
the chest multi-organ segmentation model training module 400 is configured to train a segmentation model using the fully labeled CT image dataset to obtain a chest multi-organ segmentation model;
the chest multi-organ segmentation module 500 is configured to segment chest CT image data of an organ that is not fully labeled using the chest multi-organ segmentation model.
As an embodiment, the data acquisition processing module 100 acquires CT images of the thoracic organ markers and then performs preprocessing; wherein the preprocessing includes image cropping, resampling, normalization and data enhancement.
As an implementation manner, the segmentation model adopts a STUNet model as a basic model;
the STUNet model sequentially comprises a stem block, 13 residual blocks and a seg head block to form a U-shaped structure, wherein a downsampling process is carried out among the first 6 residual blocks to form an encoder, an upsampling process is carried out among the last 6 residual blocks to form a decoder, and the residual blocks in the encoder are correspondingly connected with the residual blocks in the decoder in an end-to-end manner;
the step block comprises two processing branches, wherein one processing branch is convolution processing of 1 multiplied by 1, the other processing branch is convolution processing of two pieces of 3 multiplied by 3, and the two processing branches are combined and then used as output of the step block;
the residual block comprises two processing branches, wherein one processing branch is subjected to convolution processing of two 3 multiplied by 3 and is combined with the other processing branch to be used as output of the residual block;
the seg head block is output after being subjected to convolution processing of 1 multiplied by 1;
the downsampling process comprises two processing branches, wherein one processing branch is convolution processing with the step length of 2 being 1 multiplied by 1, and the other processing branch is convolution processing with the step length of 2 being 3 multiplied by 3 and convolution processing with the step length of 2 being 3 multiplied by 3, and the two processing branches are combined and then used as output of the downsampling process;
the up-sampling process includes nearest neighbor interpolation processing and 1×1 with step size of 1 convolution processing of x 1.
As an embodiment, the different segmentation models have a depth of (2, 2) and a width of (64, 128, 256, 512, 1024, 1024); the chest multi-organ segmentation model has a depth of (1, 1) and a width of (32, 64, 128, 256, 512, 512).
An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the method when the computer program is executed.
The electronic device may include a processing means (e.g., a central processing unit, a graphic processor, etc.) that can perform various appropriate actions and processes according to a program stored in a ROM (read only memory) or a program loaded from a storage means into a RAM (random access memory). In the RAM, various programs and data required for the operation of the electronic device 800 are also stored. The processing device, ROM and RAM are connected to each other via a bus. An I/O (input/output) interface is also connected to the bus. In general, the following devices may be connected to the I/O interface: input devices including, for example, touch screens, touch pads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices including, for example, liquid Crystal Displays (LCDs), speakers, vibrators, etc.; storage devices including, for example, magnetic tape, hard disk, etc.; a communication device. The communication means may allow the electronic device to communicate with other devices wirelessly or by wire to exchange data.
Some embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable medium, the computer program comprising program code for performing the method shown in the flow chart. In some such embodiments, the computer program may be downloaded and installed from a network via a communications device, or installed from a memory device, or installed from a ROM. The above-described functions defined in the methods of some embodiments of the present disclosure are performed when the computer program is executed by a processing device.
Some embodiments of the present disclosure may be described above as a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. The computer readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples of the computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In some embodiments of the present disclosure, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.
Finally, it is noted that the above-mentioned embodiments illustrate rather than limit the invention, and that those skilled in the art will be understood that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (10)
1. A method of chest multi-organ segmentation, comprising:
collecting CT images of the chest organ marks, dividing the CT images of the same marked organ into the same group of data sets, and sorting according to the number of the marked organs to obtain CT image data sets of a plurality of chest organ marks;
training a first segmentation model by adopting a CT image data set with the most thoracic organ marks, predicting a data set with the number of marked organs being less than that of the CT image data set with the most thoracic organ marks by adopting the trained first segmentation model, and endowing the data set with pseudo tags of the uniquely marked organs of the CT image data set with the most thoracic organ marks by adopting a prediction result;
the CT image dataset with the most thoracic organ marks and the CT image dataset with the second most thoracic organ marks are fused to train a segmentation model II, then the trained segmentation model II is adopted to predict the datasets with the number of marked organs being less than that of the CT image dataset with the second most thoracic organ marks, and the predicting result is adopted to endow the datasets with pseudo labels of the CT image dataset with the most thoracic organ marks and the CT image dataset with the second most thoracic organ marks together and uniquely mark the organs;
continuing to fuse the data sets to train the segmentation model to obtain different segmentation models and CT image data sets of a plurality of chest organ full markers corresponding to the CT image data sets of a plurality of chest organ markers;
applying pseudo labels to chest CT images of the newly incorporated organs without full markers by adopting the different segmentation models, and combining the pseudo labels with CT image data sets of the full markers of the plurality of chest organs to form a CT image data set of the full markers;
training a segmentation model by adopting the fully-marked CT image data set to obtain a chest multi-organ segmentation model;
and segmenting the chest CT image data of the organs which are not fully marked by adopting the chest multi-organ segmentation model.
2. The chest multiple organ segmentation method according to claim 1, wherein the CT image of the chest organ marker is acquired and then preprocessed; wherein the preprocessing includes image cropping, resampling, normalization and data enhancement.
3. The chest multiple organ segmentation method according to claim 1, wherein the segmentation model adopts a STUNet model as a base model;
the STUNet model sequentially comprises a stem block, 13 residual blocks and a seg head block to form a U-shaped structure, wherein a downsampling process is carried out among the first 6 residual blocks to form an encoder, an upsampling process is carried out among the last 6 residual blocks to form a decoder, and the residual blocks in the encoder are correspondingly connected with the residual blocks in the decoder in an end-to-end manner;
the step block comprises two processing branches, wherein one processing branch is convolution processing of 1 multiplied by 1, the other processing branch is convolution processing of two pieces of 3 multiplied by 3, and the two processing branches are combined and then used as output of the step block;
the residual block comprises two processing branches, wherein one processing branch is subjected to convolution processing of two 3 multiplied by 3 and is combined with the other processing branch to be used as output of the residual block;
the seg head block is output after being subjected to convolution processing of 1 multiplied by 1;
the downsampling process comprises two processing branches, wherein one processing branch is convolution processing with the step length of 2 being 1 multiplied by 1, and the other processing branch is convolution processing with the step length of 2 being 3 multiplied by 3 and convolution processing with the step length of 2 being 3 multiplied by 3, and the two processing branches are combined and then used as output of the downsampling process;
the up-sampling process includes nearest neighbor interpolation processing and 1×1 with step size of 1 convolution processing of x 1.
4. A chest multiple organ segmentation method according to claim 3, wherein the different segmentation models have a depth of (2, 2) and a width of (64, 128, 256, 512, 1024, 1024); the chest multi-organ segmentation model has a depth of (1, 1) and a width of (32, 64, 128, 256, 512, 512).
5. The chest multi-organ segmentation system is characterized by comprising a data acquisition processing module, a model training and pseudo-label giving module, a full-label data set processing module, a chest multi-organ segmentation model training module and a chest multi-organ segmentation module;
the data acquisition processing module is used for acquiring CT images of the chest organ marks, dividing the CT images of the same marked organ into the same group of data sets, and sorting the CT images according to the number of the marked organs to obtain CT image data sets of a plurality of chest organ marks;
the model training and pseudo-label giving module is used for training a segmentation model I by adopting a CT image data set with the most chest organ marks, predicting a data set with the number of marked organs being less than that of the CT image data set with the most chest organ marks by adopting the trained segmentation model I, and giving a pseudo-label of a uniquely marked organ to the CT image data set with the most chest organ marks by adopting a prediction result;
the CT image dataset with the most thoracic organ marks and the CT image dataset with the second most thoracic organ marks are fused to train a segmentation model II, then the trained segmentation model II is adopted to predict the datasets with the number of marked organs being less than that of the CT image dataset with the second most thoracic organ marks, and the predicting result is adopted to endow the datasets with pseudo labels of the CT image dataset with the most thoracic organ marks and the CT image dataset with the second most thoracic organ marks together and uniquely mark the organs;
continuing to fuse the data sets to train the segmentation model to obtain different segmentation models and CT image data sets of a plurality of chest organ full markers corresponding to the CT image data sets of a plurality of chest organ markers;
the full-label data set processing module is used for endowing pseudo labels to chest CT images of the newly-incorporated organs without full labels by adopting the different segmentation models, and combining the pseudo labels with CT image data sets of the full labels of the chest organs to form a full-label CT image data set;
the chest multi-organ segmentation model training module is used for training a segmentation model by adopting the fully-marked CT image data set to obtain a chest multi-organ segmentation model;
the chest multi-organ segmentation module is used for segmenting chest CT image data of the organs which are not fully marked by adopting the chest multi-organ segmentation model.
6. The chest multiple organ segmentation system according to claim 5, wherein the data acquisition processing module acquires CT images of the chest organ markers and then performs a pre-processing; wherein the preprocessing includes image cropping, resampling, normalization and data enhancement.
7. The chest multiple organ segmentation system according to claim 5, wherein the segmentation model employs a STUNet model as a base model;
the STUNet model sequentially comprises a stem block, 13 residual blocks and a seg head block to form a U-shaped structure, wherein a downsampling process is carried out among the first 6 residual blocks to form an encoder, an upsampling process is carried out among the last 6 residual blocks to form a decoder, and the residual blocks in the encoder are correspondingly connected with the residual blocks in the decoder in an end-to-end manner;
the step block comprises two processing branches, wherein one processing branch is convolution processing of 1 multiplied by 1, the other processing branch is convolution processing of two pieces of 3 multiplied by 3, and the two processing branches are combined and then used as output of the step block;
the residual block comprises two processing branches, wherein one processing branch is subjected to convolution processing of two 3 multiplied by 3 and is combined with the other processing branch to be used as output of the residual block;
the seg head block is output after being subjected to convolution processing of 1 multiplied by 1;
the downsampling process comprises two processing branches, wherein one processing branch is convolution processing with the step length of 2 being 1 multiplied by 1, and the other processing branch is convolution processing with the step length of 2 being 3 multiplied by 3 and convolution processing with the step length of 2 being 3 multiplied by 3, and the two processing branches are combined and then used as output of the downsampling process;
the up-sampling process includes nearest neighbor interpolation processing and 1×1 with step size of 1 convolution processing of x 1.
8. The chest multiple organ segmentation system according to claim 7, wherein the different segmentation models have a depth of (2, 2) and a width (64, 128, 256, 512, 1024, 1024); the chest multi-organ segmentation model has a depth of (1, 1) and a width of (32, 64, 128, 256, 512, 512).
9. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any of claims 1 to 4 when the computer program is executed.
10. A computer readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the steps of the method according to any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410128271.9A CN117649418B (en) | 2024-01-30 | 2024-01-30 | Chest multi-organ segmentation method and system and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410128271.9A CN117649418B (en) | 2024-01-30 | 2024-01-30 | Chest multi-organ segmentation method and system and computer readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117649418A true CN117649418A (en) | 2024-03-05 |
| CN117649418B CN117649418B (en) | 2024-04-19 |
Family
ID=90046489
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410128271.9A Active CN117649418B (en) | 2024-01-30 | 2024-01-30 | Chest multi-organ segmentation method and system and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117649418B (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110874842A (en) * | 2019-10-10 | 2020-03-10 | 浙江大学 | Chest cavity multi-organ segmentation method based on cascade residual full convolution network |
| CN110992373A (en) * | 2019-11-25 | 2020-04-10 | 杭州电子科技大学 | Deep learning-based thoracic organ segmentation method |
| US20200286614A1 (en) * | 2017-09-08 | 2020-09-10 | The General Hospital Corporation | A system and method for automated labeling and annotating unstructured medical datasets |
| AU2020103905A4 (en) * | 2020-12-04 | 2021-02-11 | Chongqing Normal University | Unsupervised cross-domain self-adaptive medical image segmentation method based on deep adversarial learning |
| US20210303930A1 (en) * | 2020-03-31 | 2021-09-30 | International Business Machines Corporation | Model training using fully and partially-annotated images |
| CN115082381A (en) * | 2022-05-27 | 2022-09-20 | 西安电子科技大学广州研究院 | Multi-target segmentation method and device based on U-shaped network |
| WO2022261472A1 (en) * | 2021-06-11 | 2022-12-15 | Genentech, Inc. | Hierarchical workflow for generating annotated training data for machine learning enabled image segmentation |
| US20230100179A1 (en) * | 2021-09-28 | 2023-03-30 | Varian Medical Systems, Inc. | Automated, collaborative process for ai model production |
| CN116824134A (en) * | 2023-05-18 | 2023-09-29 | 复旦大学 | Multi-organ segmentation system based on partial annotation dataset collaborative learning |
| CN117455925A (en) * | 2023-12-26 | 2024-01-26 | 杭州健培科技有限公司 | Chest multi-organ and rib segmentation method and device |
-
2024
- 2024-01-30 CN CN202410128271.9A patent/CN117649418B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200286614A1 (en) * | 2017-09-08 | 2020-09-10 | The General Hospital Corporation | A system and method for automated labeling and annotating unstructured medical datasets |
| CN110874842A (en) * | 2019-10-10 | 2020-03-10 | 浙江大学 | Chest cavity multi-organ segmentation method based on cascade residual full convolution network |
| CN110992373A (en) * | 2019-11-25 | 2020-04-10 | 杭州电子科技大学 | Deep learning-based thoracic organ segmentation method |
| US20210303930A1 (en) * | 2020-03-31 | 2021-09-30 | International Business Machines Corporation | Model training using fully and partially-annotated images |
| AU2020103905A4 (en) * | 2020-12-04 | 2021-02-11 | Chongqing Normal University | Unsupervised cross-domain self-adaptive medical image segmentation method based on deep adversarial learning |
| WO2022261472A1 (en) * | 2021-06-11 | 2022-12-15 | Genentech, Inc. | Hierarchical workflow for generating annotated training data for machine learning enabled image segmentation |
| US20230100179A1 (en) * | 2021-09-28 | 2023-03-30 | Varian Medical Systems, Inc. | Automated, collaborative process for ai model production |
| CN115082381A (en) * | 2022-05-27 | 2022-09-20 | 西安电子科技大学广州研究院 | Multi-target segmentation method and device based on U-shaped network |
| CN116824134A (en) * | 2023-05-18 | 2023-09-29 | 复旦大学 | Multi-organ segmentation system based on partial annotation dataset collaborative learning |
| CN117455925A (en) * | 2023-12-26 | 2024-01-26 | 杭州健培科技有限公司 | Chest multi-organ and rib segmentation method and device |
Non-Patent Citations (2)
| Title |
|---|
| MUHAMMAD IBRAHIM KHALIL等: "Multi-Scale Network for Thoracic Organs Segmentation", CMC-COMPUTERS MATERIALS & CONTINUA, 31 December 2022 (2022-12-31), pages 3251 - 3265 * |
| 邓杰航;吕灼荣;肖灿辉;: "胸部CT DICOM图像自动分割的研究与实现", 广东工业大学学报, no. 02, 15 June 2013 (2013-06-15), pages 41 - 45 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117649418B (en) | 2024-04-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11386557B2 (en) | Systems and methods for segmentation of intra-patient medical images | |
| CN110310281B (en) | Mask-RCNN deep learning-based pulmonary nodule detection and segmentation method in virtual medical treatment | |
| CN107798682B (en) | Image segmentation system, method, apparatus and computer-readable storage medium | |
| CN107067398B (en) | Completion method and device for missing blood vessels in three-dimensional medical model | |
| US8970581B2 (en) | System and method for interactive contouring for 3D medical images | |
| CN109801268B (en) | CT radiography image renal artery segmentation method based on three-dimensional convolution neural network | |
| CN112150472A (en) | Three-dimensional jaw bone image segmentation method and device based on CBCT (cone beam computed tomography) and terminal equipment | |
| CN113646653A (en) | Brain image processing | |
| CN113223028A (en) | Multi-modal liver tumor segmentation method based on MR and CT | |
| CN111210431A (en) | Blood vessel segmentation method, device, equipment and storage medium | |
| CN112561917A (en) | Image segmentation method and system, electronic device and readable storage medium | |
| CN114648525B (en) | Organ segmentation method and system based on watershed analysis | |
| CN114723893A (en) | Organ tissue spatial relationship rendering method and system based on medical images | |
| CN111292289B (en) | CT lung tumor segmentation method, device, equipment and medium based on segmentation network | |
| CN115330817A (en) | Pneumothorax lesion segmentation method based on Swin-Unet and morphological processing and related equipment | |
| WO2021030995A1 (en) | Inferior vena cava image analysis method and product based on vrds ai | |
| CN110992310A (en) | Method and device for determining partition where mediastinal lymph node is located | |
| CN111243082A (en) | Method, system, device and storage medium for obtaining digital image reconstruction image | |
| CN114419083A (en) | ResUnet medical image segmentation system based on edge operator improvement | |
| CN112634284B (en) | Weight map loss-based staged neural network CT organ segmentation method and system | |
| CN117649418B (en) | Chest multi-organ segmentation method and system and computer readable storage medium | |
| CN114446480A (en) | Mammary gland phantom generation method, device, equipment and medium | |
| CN111899273A (en) | Image segmentation method, computer device and storage medium | |
| CN112132189B (en) | CBCT image-oriented density peak superpixel preprocessing method | |
| CN115049709A (en) | Deep learning point cloud lumbar registration method for spinal minimally invasive surgery navigation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |