Method and system for automatically evaluating corruption degree of corpse CT image, and medium

Abstract

The invention provides a method and system for automatically evaluating the corruption degree of a corpse CT image and a medium, and the method comprises the following steps: obtaining the CT image of a corpse, the corruption degree corresponding to the CT image and the segmentation result of the CT image, processing the CT image, and obtaining the processed CT image; extracting multi-dimensional features from the segmentation result, wherein the multi-dimensional features include but are not limited to morphological features, pneumatosis / hydrops related features and skeleton related features; constructing an automatic classification model by using the processed CT image and the multi-dimensional features, and obtaining a weight file after training model convergence; and inputting the weight file into the processed CT image and the standardized text prompt, automatically obtaining the decay degree of the corpse, effectively utilizing the segmentation result to automatically quantify and evaluate the decay degree of the corpse, and improving the precision of the evaluation result.

Description

Technical Field

[0001] This invention relates to the field of medical imaging equipment, and more particularly to a method, system, and medium for automatically assessing the degree of decomposition in cadaver CT images. Background Technology

[0002] Assessing the degree of decomposition of a corpse is one of the core tasks of forensic pathology, and it is of great significance for forensic identification practices and scientific research, such as estimating the time of death, analyzing the cause of death, and determining the manner of death. Traditional assessment methods rely on forensic experts' experience in judging morphological changes on the surface of the body and internal organs, which has problems such as strong subjectivity, many interfering factors, and insufficient standardization.

[0003] With the maturation of medical imaging technology, cadaver CT scans have become an important means of obtaining information about the internal structure of cadavers. While the rapid development of deep learning technology has enabled automated segmentation and extraction of major organs, fluid accumulations, gas accumulations, and bones throughout the human body, current technologies have failed to effectively integrate segmentation results with the assessment of the degree of putrefaction. The lack of quantitative models hinders the full realization of the application value of segmentation data. Therefore, a solution for objectively and accurately assessing the degree of putrefaction is urgently needed. Summary of the Invention

[0004] In order to overcome the above-mentioned technical defects, the purpose of this invention is to provide a method, system and medium for automatically assessing the degree of decomposition of cadaver CT images.

[0005] This invention discloses a method for automatically assessing the degree of decomposition in cadaver CT images, comprising the following steps: The CT images of the cadaver, the degree of decomposition corresponding to the CT images, and the segmentation results of the CT images are obtained. The CT images are then processed to obtain the processed CT images. Multi-dimensional features are extracted from the segmentation result, including but not limited to morphological features, air / fluid accumulation-related features, and skeletal-related features; Using the processed CT image and the multi-dimensional features, an automatic classification model is constructed, and a weight file is obtained after the model is trained and converged. By inputting the weighted file into the processed CT image and the standardized text prompt, the degree of decomposition of the corpse can be automatically obtained.

[0006] Preferably, the process of processing the CT image to obtain the processed CT image includes the following steps: The CT image value is linearly transformed according to formula (1) and the result is mapped to the range [0,1], where V is the value of the CT image; (1) Incremental data processing is performed on the CT image, including but not limited to flipping, rotating, and cropping.

[0007] Preferably, the extraction of multi-dimensional features from the segmentation result image includes: Extract morphological features from the segmentation results, including but not limited to organ volume variation features and organ shape irregularity features; Among them, the organ volume is calculated by segmenting the three-dimensional model of the organ, and compared with the corresponding organ volume database of normal human body to obtain the volume change rate, and then obtain the organ volume change characteristics. Using a contour extraction algorithm, the contours of segmented organs are extracted, and the shape descriptors of the organs are calculated. These shape descriptors include, but are not limited to, the ratio of the organ's perimeter to its area, thus obtaining the irregular shape features of the organs.

[0008] Preferably, the extraction of multi-dimensional features from the segmentation result image includes: Extract gas / liquid related features from the segmentation results, including but not limited to gas distribution features and liquid density and quantity features; Among them, based on the gas accumulation area of ​​the segmentation result, the distribution ratio of the gas accumulation area in different cavities is statistically analyzed to obtain the gas accumulation distribution characteristics; The average CT value of the fluid accumulation area in the segmentation result is measured, and the volume of the fluid accumulation is calculated to obtain the characteristics of the fluid density and quantity.

[0009] Preferably, the extraction of multi-dimensional features from the segmentation result image includes: Extract bone-related features from the segmentation results, including but not limited to statistical changes in the density of soft tissue surrounding the bones; In this segmentation result, a region of interest of a specific thickness is defined around the bone region, and the average CT value of the soft tissue within this region of interest is calculated. and the standard deviation of CT values Let N be the number of pixels within the soft tissue ROI surrounding the bone, and let CTi be the CT value of each pixel. The average CT value CT is calculated according to formulas (4-1) and (4-2). and standard deviation This allows for the acquisition of changes in the density of soft tissues surrounding the bones. (4-1) (4-2) Preferably, the automatic classification model is constructed using the processed CT image and the multi-dimensional features, and a weight file is obtained after the model is trained and converged, including the following steps: The processed CT image is image encoded to obtain image features, and these multi-dimensional features are then text encoded to obtain text features. The text features and image features are combined using a linear layer to classify the degree of corruption, forming an automatic classification model based on deep learning.

[0010] Preferably, the multi-dimensional features are encoded into text features to obtain text features, including the following steps: Based on this multi-dimensional feature, a standardized text prompt is automatically generated using a given word or phrase as a prompt and a fixed template s=['text prompt'].

[0011] Preferably, the CT image of the corpse is a three-dimensional medical image of the corpse obtained by CT scan. The degree of decomposition corresponding to the CT image includes, but is not limited to, mild, moderate and severe. The segmentation result of the CT image includes, but is not limited to, the segmentation result images of major organs, bones, fluid accumulation, gas accumulation, etc. in the CT image.

[0012] This invention also discloses a system for automatically assessing the degree of decomposition in cadaver CT images, comprising: The image processing module is used to acquire CT images of the cadaver, the degree of decomposition corresponding to the CT image, and the segmentation results of the CT image, and to process the CT image to obtain the processed CT image. The multi-dimensional feature extraction module is used to extract multi-dimensional features from the segmentation result. These multi-dimensional features include, but are not limited to, morphological features, air / fluid accumulation-related features, and bone-related features. The network model training module is used to build an automatic classification model using the processed CT image and the multi-dimensional features. After the training model converges, a weight file is obtained. The network model inference module is used to input the weight file into the processed CT image and the standardized text prompt to automatically obtain the degree of decomposition of the corpse.

[0013] The present invention also discloses a computer-readable storage medium having a computer program stored thereon, the computer executing a method for automatically assessing the degree of decomposition of any of the cadaver CT images.

[0014] Compared with existing technologies, the above technical solution has the following advantages: 1. Multi-dimensional feature fusion: Breaking through the limitations of traditional single features (such as external observation of the corpse), it integrates multi-tissue and multi-parameter features such as organs, gas accumulation, fluid accumulation, and bones to comprehensively and systematically reflect the putrefaction process; 2. Automatically generate standardized text prompts from multi-dimensional features, and combine them with image features from the segmentation results to construct a corruption assessment model; 3. End-to-end automation: The degree of decomposition is directly output from the cadaver CT images and segmentation results without human intervention, significantly improving the efficiency of forensic work. Attached Figure Description

[0015] Figure 1 This is a flowchart of the steps of a method for automatically assessing the degree of decomposition in cadaver CT images disclosed in this invention; Figure 2 This is a structural framework diagram of a system for automatically assessing the degree of decomposition in cadaver CT images disclosed in this invention; Figure 3 This is a structural framework diagram of the image processing module disclosed in a preferred embodiment of the present invention; Figure 4 This is a structural framework diagram of the multi-dimensional feature extraction module disclosed in a preferred embodiment of the present invention; Figure 5 This is a structural framework diagram of the network model training module disclosed in a preferred embodiment of the present invention; Figure 6 This is a functional structure diagram of the network model training module disclosed in a preferred embodiment of the present invention; Figure 7 This is a flowchart of an automatic method for assessing the degree of corruption, as disclosed in a preferred embodiment of the present invention.

[0016] Figure labeling: 100 - Image processing module; 101 - Normalized linear transformation module; 102 - Data augmentation module; 200 - Multi-dimensional feature extraction module; 201 - Morphological feature extraction module; 202 - Gas / fluid accumulation related feature extraction module; 203 - Skeletal related feature extraction module; 300 - Network model training module; 301 - Image encoder module; 302 - Text prompt encoder module; 303 - Classifier module; 400 - Network model inference module. Detailed Implementation

[0017] The advantages of the present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments.

[0018] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0019] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The singular forms “a,” “the,” and “the” as used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0020] It should be understood that although the terms first, second, third, etc., may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are used only to distinguish information of the same type from one another. For example, without departing from the scope of this disclosure, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0021] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "up", "down", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0022] In the description of this invention, unless otherwise specified and limited, it should be noted that the terms "installation", "connection" and "linking" should be interpreted broadly. For example, they can refer to mechanical or electrical connections, or internal connections between two components. They can be direct connections or indirect connections through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms according to the specific circumstances.

[0023] In the following description, suffixes such as "module," "part," or "unit" used to denote elements are used only for the convenience of the description of the invention and have no specific meaning in themselves. Therefore, "module" and "part" can be used interchangeably.

[0024] like Figure 1 As shown, to achieve the above objective, this invention discloses a method for automatically assessing the degree of decomposition in cadaver CT images, comprising the following steps: Step S10: Obtain the CT image of the corpse, the degree of decomposition corresponding to the CT image, and the segmentation result of the CT image; process the CT image to obtain the processed CT image. Step S20: Extract multi-dimensional features from the segmentation result. The multi-dimensional features include, but are not limited to, morphological features, air / fluid accumulation-related features, and bone-related features. Step S30: Using the processed CT image and the multi-dimensional features, construct an automatic classification model, and obtain a weight file after the training model converges; Step S40: Input the weight file into the processed CT image and the standardized text prompt to automatically obtain the degree of decomposition of the corpse.

[0025] Specifically, this invention is applied to the field of assessing the degree of decomposition of corpses. The disclosed method for automatically assessing the degree of decomposition is based on a deep learning model. It utilizes CT images of the corpse and segmentation results to extract quantitative features related to decomposition and train the decomposition assessment model. It should be noted that the data involved in the CT images are all raw data, stored in DICOM data format. The images involved include, but are not limited to, different causes of death, different degrees of decomposition, and whether the body is frozen. The degree of decomposition corresponding to the images is distinguished as mild, moderate, and severe. The segmented images involved can be the results of automatic, semi-automatic, or interactive segmentation, and the segmentation method is not limited and is existing technology. The major organs involved include common tissues and organs (heart, liver, lungs, etc.), and the skeleton is detailed down to each rib and each vertebra.

[0026] In one embodiment of the present invention, the above method further includes: The degree of decomposition corresponding to the CT image and the segmentation result of the CT image are compared with the degree of decomposition of the corpse automatically output by the automatic classification model to verify the accuracy of the automatic classification model.

[0027] Specifically, the segmentation results in the training data are labeled by the comprehensive judgment of multiple senior forensic experts, and the degree of corruption is also determined by the comprehensive judgment of multiple senior forensic experts, ensuring the authority of the segmentation results and the degree of corruption. The model output results are compared with the segmentation results and the comprehensive label of corruption degree determined by human judgment to verify the accuracy of the assessment. Furthermore, in step S10, the CT image of the corpse is a three-dimensional medical image of the corpse obtained by CT scan. The degree of decomposition corresponding to the CT image includes, but is not limited to, mild, moderate and severe. The segmentation result of the CT image includes, but is not limited to, the segmentation result images of major organs, bones, fluid accumulation, gas accumulation and so on in the CT image.

[0028] Further, in step S10, the CT image is processed to obtain the processed CT image, including the following steps: According to formula (1), the value of the CT image is linearly transformed to map the result to the range of [0,1], where V is the CT value; (1) Incremental data processing is performed on the CT image, including but not limited to flipping, rotating, and cropping.

[0029] Specifically, CT image processing mainly involves normalizing the CT image using a linear transformation and performing data augmentation on both the CT image and the labeled image. Normalization is done to facilitate convergence during subsequent network training, while data augmentation allows limited data to generate value equivalent to more data without substantially increasing the amount of data.

[0030] Further, in step S20, multi-dimensional features are extracted from the segmentation result image, including: Extract morphological features from the segmentation results, including but not limited to organ volume variation features and organ shape irregularity features; Among them, the organ volume is calculated by segmenting the three-dimensional model of the organ, and compared with the corresponding organ volume database of normal human body to obtain the volume change rate, and then obtain the organ volume change characteristics. Specifically, the volume of organs (such as liver, heart, lungs, etc.) is calculated using automatically segmented three-dimensional models. The volume is then compared with a database of corresponding organ volumes in a normal human body to obtain the volume change rate. For example, the formula for calculating the liver volume change rate Rliver is shown in formula (2). (2) in, The volume of the liver obtained from the segmentation of the corpse. This represents the average volume of a normal human liver; the calculation methods for the volume change rates of other major organs are approximate.

[0031] Using a contour extraction algorithm, the contours of segmented organs are extracted, and the shape descriptors of the organs are calculated. These shape descriptors include, but are not limited to, the ratio of the organ's perimeter to its area, thus obtaining the irregular features of the organ's shape. Specifically, contour extraction algorithms are used to extract the contours of segmented organs and calculate shape descriptors such as the ratio of their perimeter to their area to reflect the degree of irregularity in the organ's shape. Taking the liver as an example, let the perimeter of the liver contour be C and the area be A, then the formula for calculating the shape irregularity Sliver is shown in formula (3). (3) During the putrefaction process, autolysis of liver tissue usually leads to increased irregularity in shape, while the calculation method for the volume change rate of other major organs is approximately the same.

[0032] In this embodiment of the invention, extracting multi-dimensional features further includes: extracting gas / liquid related features of the segmentation results, including but not limited to gas distribution features and liquid density and quantity features; Among them, based on the gas accumulation area of ​​the segmentation result, the distribution ratio of the gas accumulation area in different cavities is statistically analyzed to obtain the gas accumulation distribution characteristics; Specifically, for the segmented areas of air accumulation, their distribution ratio in different body cavities (thoracic cavity, abdominal and pelvic cavities, etc.) is statistically analyzed. For example, the proportion of air accumulation in the thoracic cavity... Volume of pneumothorax With total volume of gas The ratio of pneumothorax to pneumothorax is similarly calculated. In one embodiment of the invention, pneumothorax located at the upper edge of the first rib, at the level of the diaphragm, and at the inner edge of the ribs is considered as pleural effusion, while pneumothorax located at the level of the diaphragm, at the lower edge of the fifth lumbar vertebra, and at the inner edge of the lumbar vertebrae and pelvis is considered as abdominopelvic effusion. Here, based on the organ segmentation results, the junction of the lung and liver is marked at the level of the diaphragm. Typically, in the early stages of putrefaction, pneumothorax may first appear in specific locations such as the intestines, and its distribution will change as putrefaction progresses.

[0033] The average CT value of the fluid accumulation region in the segmentation result is measured, and the volume of the fluid accumulation is calculated to obtain the characteristics of the fluid density and quantity. Specifically, the average CT value of the segmented fluid accumulation region is measured, and the volume of the fluid is calculated simultaneously. The average CT value is the average of the CT values ​​of all pixels in the fluid accumulation region, and the fluid volume is calculated using the number of pixels in the segmented region and the voxel volume. The CT value of hematoma is generally higher than other types of fluid accumulation, and the fluid volume changes as putrefaction progresses.

[0034] In this embodiment of the invention, extracting multi-dimensional features further includes: Extract bone-related features from the segmentation results, including but not limited to statistical changes in the density of soft tissue surrounding the bones; In this segmentation result, a region of interest of a specific thickness is defined around the bone region, and the average CT value of the soft tissue within this region of interest is calculated. and the standard deviation of CT values Let N be the number of pixels within the soft tissue ROI surrounding the bone, and let CTi be the CT value of each pixel. The average CT value CT is calculated according to formulas (4-1) and (4-2). and standard deviation This allows for the acquisition of changes in the density of soft tissues surrounding the bones; (4-1) (4-2) Specifically, as a corpse decomposes, the soft tissues surrounding the bones liquefy and decompose, leading to a decrease in their average CT value and an increase in their standard deviation.

[0035] Furthermore, step S30 includes the following steps: The processed CT image is image encoded to obtain image features, and these multi-dimensional features are then text encoded to obtain text features. The text features and image features are combined using a linear layer to classify the degree of corruption, forming an automatic classification model based on deep learning.

[0036] In one embodiment of the present invention, the multi-dimensional features are encoded into text features to obtain text features, including the following steps: Based on this multi-dimensional feature, a standardized text prompt is automatically generated using a given word or phrase as a prompt and a fixed template s=['text prompt'].

[0037] Example 1: like Figure 2 As shown, a preferred embodiment of the present invention includes the following steps: Step S51: Acquire three-dimensional medical images of the cadaver obtained from CT scans. All data is raw and stored in DICOM format. The images cover different causes of death, different degrees of decomposition, whether frozen, etc. Step S52: Obtain the degree of decomposition corresponding to the image (distinguished as mild, moderate, and severe). The degree of decomposition classification involved is obtained by combining the comprehensive assessment results of multiple professional forensic experts.

[0038] Step S53: Obtain segmentation results images of major organs, bones, fluid accumulation, and gas accumulation from the CT images. The segmentation images involved can be automatically segmented, semi-automatically segmented, or interactively segmented, and the segmentation method is not limited. Major organs include common tissues and organs (heart, liver, lungs, etc.), and bones are refined to each rib, each vertebra, etc.

[0039] Step S54: Perform a normalized linear transformation on the CT image to map the result to the range [0,1], which facilitates convergence during subsequent network training; perform data augmentation operations on the CT image and the labeled image to increase the diversity of data samples.

[0040] Step S55 involves extracting different features from each segmentation result in S3, including automatic extraction of organ morphological features, gas / fluid accumulation-related features, and bone-related features.

[0041] Step S56: Automatically generate standardized text prompts for the multi-dimensional features obtained in S5 according to a fixed template; Step S57: Construct an automatic classification model for the degree of corruption that combines image features and text features until the training model converges, obtain the weights of the entire network, and save them.

[0042] Step S58: Load the trained weight file, input the cadaver CT image and standardized text prompts processed by steps S4, S5 and S6, and automatically obtain the degree of decomposition (such as mild, moderate, severe, etc.).

[0043] like Figure 3 As shown, to achieve the above objectives, the present invention also discloses a system for automatically assessing the degree of decomposition in cadaver CT images, comprising: The image processing module 100 is used to acquire CT images of the corpse, the degree of decomposition corresponding to the CT images, and the segmentation results of the CT images, and to process the CT images to obtain processed CT images. The multi-dimensional feature extraction module 200 is used to extract multi-dimensional features from the segmentation result. These multi-dimensional features include, but are not limited to, morphological features, air / fluid accumulation-related features, and bone-related features. The network model training module 300 is used to construct an automatic classification model using the processed CT image and the multi-dimensional features. After the training model converges, a weight file is obtained. The network model inference module 400 is used to input the weight file into the processed CT image and the standardized text prompt to automatically obtain the degree of decomposition of the corpse.

[0044] Furthermore, such as Figure 3 As shown, the image processing module 100 includes a normalized linear transformation module 101 and a data augmentation module 102; The normalized linear transformation module 101 is used to perform a linear transformation on the CT value v of the input image, so that the result is mapped to the range of [0,1], which facilitates convergence in the subsequent network training process. The normalization formula is shown in formula (1): (1) The data augmentation module 102 is used to make limited data generate value equivalent to more data without substantially increasing the data; common data augmentation methods include flipping, rotating, and cropping.

[0045] Furthermore, such as Figure 4 As shown, the multi-dimensional feature extraction module 200 includes a morphological feature extraction module 201, an air / fluid accumulation related feature extraction module 202, and a skeleton related feature extraction module 203.

[0046] Among them, the morphological feature extraction module 201 is used to extract organ volume change features and organ shape irregularity features.

[0047] Organ volume change characteristics: The volume of organs (such as liver, heart, lungs, etc.) is calculated using automatically segmented three-dimensional models. The volume change rate is obtained by comparing the models with a database of corresponding organ volumes in normal human bodies. For example, the formula for calculating the liver volume change rate Rliver is shown in formula (2). (2) in, The volume of the liver obtained from the segmentation of the corpse. This represents the average volume of a normal human liver; the calculation methods for the volume change rates of other major organs are approximate.

[0048] Organ shape irregularity characteristics: Using contour extraction algorithms, the contours of segmented organs are extracted, and shape descriptors such as the ratio of their perimeter to their area are calculated to reflect the degree of irregularity of the organ shape. Taking the liver as an example, let the perimeter of the liver contour be C and the area be A, then the formula for calculating the shape irregularity Sliver is shown in formula (3). (3) During the putrefaction process, autolysis of liver tissue usually leads to increased irregularity in shape, while the calculation method for the volume change rate of other major organs is approximately the same.

[0049] The gas / liquid related feature extraction module 202 is used to extract the gas distribution characteristics and the density and amount of liquid; Pneumothorax distribution characteristics: For the segmented pneumothorax areas, the distribution ratio of these areas in different body cavities (thoracic cavity, abdominal and pelvic cavities, etc.) is statistically analyzed. For example, the proportion of pneumothorax in the thoracic cavity. Volume of pneumothorax With total volume of gas The ratio of pneumothorax to pneumothorax is similarly calculated. Here, pneumothorax located at the upper edge of the first rib, at the level of the diaphragm, and at the inner edge of the ribs is considered pleural effusion, while pneumothorax located at the level of the diaphragm, at the lower edge of the fifth lumbar vertebra, and at the inner edge of the lumbar vertebrae and pelvis is considered abdominopelvic effusion. Here, based on the organ segmentation results, the junction of the lung and liver is marked at the level of the diaphragm. Typically, in the early stages of putrefaction, pneumothorax may first appear in specific locations such as the intestines, and its distribution will change as putrefaction progresses.

[0050] Characteristics of fluid density and volume: The average CT value of the segmented fluid accumulation region is measured, and the volume of the fluid is calculated simultaneously. The average CT value is the average of the CT values ​​of all pixels in the fluid accumulation region, and the fluid volume is calculated using the number of pixels in the segmented region and the voxel volume. The CT value of hematoma is generally higher than other types of fluid accumulation, and the fluid volume changes with the progression of putrefaction.

[0051] The skeletal feature extraction module 203 is used to statistically analyze changes in the density of soft tissues surrounding the bones. A region of interest (ROI) of a certain thickness is defined around the segmented skeletal region. The average CT value and standard deviation of the CT value within this ROI are calculated. As the body decomposes, the soft tissue surrounding the bones undergoes liquefaction and decomposition, leading to a decrease in the average CT value and an increase in the standard deviation. Let N be the number of pixels within the ROI of soft tissue surrounding the bones, and let CTi be the CT value of each pixel. Then, the average CT value CT... and standard deviation The calculation formula is shown in formula (4): (4-1) (4-2) The network model training module 300 includes an image encoder module 301, a text prompt encoder module 302, and a classifier module 303; In the image encoder module 301, the input image of the image encoder is the CT image processed by the image processing module. The encoder can be ViT, ResNet, or other backbones. Here, ViT is used as an example. The input CT image is divided into a set of patches of size (PD×PH×PW). The vectorized xp is mapped to the potential D-dimensional embedding space using trainable linear projection. The result after mapping is called a token, and its size is (N,D). At the same time, a learnable embedding vector Class Token (vector is 0) is added for subsequent classification prediction. At this time, the size of all tokens is (N+1,D). In order to encode the spatial information of the patch, a specific position embedding is embedded in the token to retain the position information, thereby adding position information to each path, as shown in formula (5): (5) in, For mapping projection, The embedding location is shared and learnable across all images. The encoder primarily consists of an L-layer multi-head self-attention (MSA) mechanism and a multi-layer perceptron (MLP) module. Therefore, the output of the L-layer is as follows: (6) (7) Where LN(·) represents the normalization operation, This represents the encoded image, where the output of the image encoder is the image embedded in Zimage.

[0052] In the text prompt encoder module 302, the input to the text prompt encoder is the automatically generated standardized text prompt from the features obtained by the multi-dimensional feature extraction module. This is accomplished by providing a word or phrase as a prompt, using a fixed template s=['text prompt']. For ease of understanding, an example is given below: ['Liver volume change rate is 10%, shape irregularity is 4, pleural effusion ratio is 0%, abdominal effusion ratio is 7%, effusion density is 40, effusion volume is 400, mean value of soft tissue around bones is 60, standard deviation of soft tissue around bones is 5']. This example can include the volume change rate and shape irregularity of multiple organs, and the effusion density and volume can also be 0. The text encoder in the CLIP model can be used directly to encode the input text prompt. The text encoder accepts s as input and outputs the text embedding Ztext.

[0053] The classifier module 303 is used to input the image embedding Zimage and the text embedding Ztext together into the classifier module to complete the final classification of the degree of corruption. Here, the classifier only needs to use a single linear layer. We simply divide the degree of corruption into mild, moderate, and severe, or we can use a scoring mechanism (0~10, etc.).

[0054] The network model inference module 400 loads the trained weight file into the deep learning network model, takes the CT image and the segmentation results as input, and outputs the degree of corruption. The input CT image has passed through the image processing module, and the segmentation results images have passed through the multi-dimensional feature extraction module and generated a fixed template s=['text prompt'] format. Figure 5 As shown, this process directly outputs the degree of decomposition from the cadaver's CT images and segmentation results, requiring no manual intervention and significantly improving the efficiency of forensic work.

[0055] It should be noted that the system for automatically assessing the degree of decomposition in cadaver CT images corresponds to the method described above; for other unexplained parts, please refer to the content in the method section.

[0056] To achieve the above objectives, the present invention also provides a computer-readable storage medium comprising multiple storage media, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, server, application store, etc., wherein a computer program is stored thereon, and the program performs corresponding functions when executed by a processor. The computer-readable storage medium of this embodiment is used to execute any of the above-described methods for automatically assessing the degree of decomposition of cadaver CT images.

[0057] One embodiment of the present invention is based on a deep learning model, which uses cadaver CT images and segmentation results (organs, fluid accumulation, gas accumulation, bones, etc.) to extract putrefaction-related quantitative features, trains a putrefaction assessment model, and achieves the following effects: First, automation: the entire process requires no human intervention and directly outputs the degree of decomposition from the cadaver CT images; Second, objectivity: Based on quantitative characteristics and model evaluation, the influence of subjective experience is eliminated; Third, high adaptability: suitable for corpses in complex scenarios with different degrees of decomposition (mild to high) and different causes of death; Fourth, high precision: It integrates multi-dimensional features and text prompt features to improve the accuracy of the evaluation.

[0058] It should be noted that the embodiments of the present invention have better implementability and are not intended to limit the present invention in any way. Any person skilled in the art may use the above-disclosed technical content to change or modify it into equivalent effective embodiments. However, any modifications or equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of the technical solution of the present invention.

Claims

1. A method for automatically assessing the degree of decomposition in cadaver CT images, characterized in that, Includes the following steps: The CT images of the cadaver, the degree of decomposition corresponding to the CT images, and the segmentation results of the CT images are obtained. The CT images are then processed to obtain the processed CT images. Multi-dimensional features are extracted from the segmentation results, including but not limited to morphological features, air / fluid accumulation-related features, and skeletal-related features; Using the processed CT images and the multi-dimensional features, an automatic classification model is constructed, and a weight file is obtained after the model is trained and converged. The weighted file is input into the processed CT image and the standardized text prompt to automatically obtain the degree of decomposition of the corpse.

2. The method for automatically assessing the degree of decomposition using cadaver CT images as described in claim 1, characterized in that, The process of processing CT images to obtain processed CT images includes the following steps: The values ​​of the CT image are linearly transformed according to formula (1) and the result is mapped to the range [0,1], where V is the value of the CT image; (1) Incremental data processing is performed on the CT images, including but not limited to flipping, rotating, and cropping.

3. The method for automatically assessing the degree of decomposition using cadaver CT images as described in claim 1, characterized in that, The extraction of multi-dimensional features from the segmented image includes: Extract morphological features from the segmentation results, including but not limited to organ volume variation features and organ shape irregularity features; Among them, the organ volume is calculated by segmenting the three-dimensional model of the organ, and compared with the corresponding organ volume database of normal human body to obtain the volume change rate, and then obtain the organ volume change characteristics. Using a contour extraction algorithm, the contours of segmented organs are extracted, and the shape descriptors of the organs are calculated. These shape descriptors include, but are not limited to, the ratio of the organ's perimeter to its area, thus obtaining the irregular shape features of the organs.

4. The method for automatically assessing the degree of decomposition using cadaver CT images as described in claim 1, characterized in that, The extraction of multi-dimensional features from the segmented image includes: Extract gas / liquid related features from the segmentation results, including but not limited to gas distribution features and liquid density and quantity features; Among them, based on the gas accumulation area of ​​the segmentation result, the distribution ratio of the gas accumulation area in different cavities is statistically analyzed to obtain the gas accumulation distribution characteristics; The average CT value of the fluid accumulation area in the segmentation result is measured, and the volume of the fluid accumulation is calculated to obtain the characteristics of the fluid density and quantity.

5. The method for automatically assessing the degree of decomposition using cadaver CT images as described in claim 1, characterized in that, The extraction of multi-dimensional features from the segmented image includes: Extract bone-related features from the segmentation results, including but not limited to statistical changes in the density of soft tissue surrounding the bones; In the segmentation results, a region of interest of a specific thickness is defined around the skeletal region, and the average CT value of the soft tissue within the region of interest is calculated. and the standard deviation of CT values Let N be the number of pixels within the soft tissue ROI surrounding the bone, and let CTi be the CT value of each pixel. The average CT value CT is calculated according to formulas (4-1) and (4-2). and standard deviation This allows for the acquisition of changes in the density of soft tissues surrounding the bones. （4-1） （4-2） The method for automatically assessing the degree of decomposition of cadaver CT images as described in claim 1, characterized in that the step of constructing an automatic classification model using the processed CT image and the multi-dimensional features, and obtaining a weight file after training the model to converge, includes the following steps: The processed CT image is image encoded to obtain image features, and the multi-dimensional features are text encoded to obtain text features; The text features and image features are combined using a linear layer to classify the degree of corruption, forming an automatic classification model based on deep learning.

6. The method for automatically assessing the degree of decomposition using cadaver CT images as described in claim 1, characterized in that, The process of encoding the multi-dimensional features into text features includes the following steps: Based on the aforementioned multi-dimensional features, a standardized text prompt is automatically generated using a given word or phrase as a prompt and a fixed template s=['text prompt'].

7. The method for automatically assessing the degree of decomposition using cadaver CT images as described in claim 1, characterized in that, The CT images of the corpse are three-dimensional medical images of the corpse obtained by CT scan. The degree of decomposition corresponding to the CT images includes, but is not limited to, mild, moderate and severe. The segmentation results of the CT images include, but are not limited to, the segmentation results of major organs, bones, fluid accumulation and gas accumulation in the CT images.

8. A system for automatically assessing the degree of decomposition in cadaver CT images, characterized in that, include: The image processing module is used to acquire CT images of the cadaver, the degree of decomposition corresponding to the CT images, and the segmentation results of the CT images, and to process the CT images to obtain processed CT images. A multi-dimensional feature extraction module is used to extract multi-dimensional features from the segmentation results. The multi-dimensional features include, but are not limited to, morphological features, air / fluid accumulation-related features, and bone-related features. The network model training module is used to construct an automatic classification model using the processed CT images and the multi-dimensional features, and obtain a weight file after the training model converges. The network model inference module is used to input the weight file into the processed CT image and the standardized text prompt to automatically obtain the degree of decomposition of the corpse.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer performs a method for automatically assessing the degree of decomposition from cadaver CT images according to any one of claims 1-8.

Images