Training method of cadaver injury result prediction model, injury prediction method and system

Abstract

The present disclosure provides a training method for a corpse injury result prediction model, an injury prediction method and a system, wherein the training method for the corpse injury result prediction model includes obtaining a plurality of groups of sample data; wherein any group of the sample data includes a sample image, and associated identification information corresponding to a preset injury site on the corpse in the sample image; each group of the sample images is used as the input of a target preset model, and the corresponding associated identification information is used as the output of the target preset model, so as to train and obtain the corpse injury result prediction model for predicting the corpse injury condition in any image. Through the training method for the corpse injury result prediction model provided by the present disclosure, the corpse injury result on the target image can be identified quickly and accurately, thereby improving the efficiency of identifying the corpse injury result.

Description

Technical Field

[0001] The present disclosure relates to the field of image processing, and in particular to a training method for a corpse injury result prediction model, an injury prediction method and a system. Background Art

[0002] Acceleration injury refers to the injury caused by the head being stationary and the moving object hitting the head, causing the brain to move at an accelerated speed, such as stick injuries and hand injuries; deceleration injury refers to the injury caused by the moving head hitting a stationary force point, such as most traffic accidents, falls from heights, and falls. Among them, the acceleration injury mechanism mostly only causes brain damage on the side of the head that is hit - impact injury, while the deceleration injury mechanism mostly causes brain damage on the opposite side of the head that is hit - contralateral injury. However, in forensic practice, acceleration and deceleration injuries are often judged based on whether there is an impact injury or contralateral injury, but there is a lack of practical research on acceleration and deceleration injuries in real cases.

[0003] The existing methods for identifying acceleration and deceleration craniocerebral injuries and determining the injury mechanism of corpses, for example, in the case of unclear cases, atypical injuries, and lack of effective witnesses or video evidence, mainly rely on forensic doctors to observe the corpse's craniocerebral CT (computer tomography) images under a viewing light or on a computer or tablet screen, and to determine the injury mechanism based on personal knowledge and long-term experience combined with the morphological changes of acceleration and deceleration injury images. However, the identification personnel make judgments based on their limited personal experience, which not only has uneven experience, but also may miss or even misjudge due to negligence or work fatigue.

[0004] Since a systematic study of the injury mechanism of traumatic craniocerebral injury requires a large amount of cadaver case sample data, however, the traditional autopsy rate is not high and the sample acquisition source is limited; the use of CT imaging technology to examine craniocerebral injury in the field of forensic medicine has not been carried out for a long time, and there is a lack of a large number of craniocerebral injury imaging data databases that can be used for research; deep learning as an emerging technology needs a large number of examples to verify the above work. As a result, the research on the injury mechanism of acceleration and deceleration craniocerebral injury is relatively lagging, resulting in the inability to correctly identify the injury of corpses. Summary of the invention

[0005] The technical problem to be solved by the present invention is to overcome the defect in the prior art that the injury condition of a corpse cannot be determined, and to provide a training method for a corpse injury result prediction model, an injury prediction method and a system.

[0006] The present invention solves the above technical problems through the following technical solutions:

[0007] According to a first aspect of the present disclosure, a method for training a cadaver injury outcome prediction model is provided, the training method comprising:

[0008] Acquire several groups of sample data; wherein any group of the sample data includes a sample image and sample association identification information corresponding to a preset injury site on the corpse in the sample image;

[0009] Each group of the sample images is used as the input of the target preset model, and the corresponding sample association identification information is used as the output of the target preset model to train and obtain the corpse injury result prediction model for predicting the corpse injury condition in any image.

[0010] Preferably, the step of obtaining several groups of sample data includes:

[0011] Acquire a number of the sample images;

[0012] Performing labeling processing on the sample image to obtain a first sample damage result corresponding to the sample image;

[0013] The first sample damage result includes at least one of whether there is damage, damage location information, and first damage type information;

[0014] The step of using each group of the sample images as the input of the target preset model and using the corresponding sample association identification information as the output of the target preset model to train the body injury result prediction model for predicting the body injury condition in any image includes:

[0015] Using the sample image as input of a first preset model, and using the sample injury result corresponding to the sample image as output of the first preset model, so as to train and obtain the corpse injury result prediction model for predicting corpse injury conditions in any image;

[0016] or,

[0017] The step of obtaining several groups of sample data comprises:

[0018] Acquire a number of the sample images;

[0019] Performing labeling processing on the sample image to obtain a second sample damage result corresponding to the sample image;

[0020] The second sample injury result includes at least one of whether there is an injury, injury location information, first injury type information, and second injury type information; the step of using each group of the sample images as the input of the preset model, and using the corresponding sample association identification information as the output of the preset model to train the corpse injury result prediction model for predicting the corpse injury condition in any image includes:

[0021] Using the sample image as input of a second preset model, and using a second sample injury result corresponding to the sample image as output of the second preset model, so as to train and obtain the corpse injury result prediction model for predicting the injury condition corresponding to the corpse injury position in any image;

[0022] Wherein, the first injury type information is used to characterize the acceleration and deceleration injury type of the injury on the corpse;

[0023] The second injury type information is used to characterize the condition type of the injury on the corpse.

[0024] Preferably, the first preset model includes a DeepLabv3+ model (a semantic segmentation model);

[0025] and / or,

[0026] The second preset model includes a ResNet18-ASPP model (a semantic segmentation model);

[0027] The ResNet18-ASPP model includes a first convolutional layer, a first residual structure layer, a second residual structure layer, a third residual structure layer, a fourth residual structure layer, a perforated spatial pyramid pooling structure, and an output layer, which are connected in sequence.

[0028] Preferably, before the sample image is annotated, the training method further includes:

[0029] Determining whether the sample image is an image in a preset format, and if not, converting the sample image into an image in a preset format;

[0030] and / or,

[0031] The sample images include CT images, photographed images, single scanned images, or images obtained by superimposing multiple consecutive scanned images.

[0032] Preferably, the preset injury site includes the brain;

[0033] The injury type corresponding to the first injury type information includes accelerated craniocerebral injury and / or decelerated craniocerebral injury;

[0034] The injury type corresponding to the second injury type information includes at least one of cerebral contusion, subarachnoid hemorrhage, scalp hematoma, skull fracture, subdural hematoma, epidural hematoma and cerebral hemorrhage.

[0035] According to a second aspect of the present disclosure, a method for predicting cadaver injuries is provided, the method comprising:

[0036] Get the target image;

[0037] Inputting the target image into a cadaver injury result prediction model, and outputting a target prediction result representing the injury condition on the cadaver in the target image;

[0038] Wherein, the cadaver injury outcome prediction model is obtained by using the training method as described in the first aspect of the present disclosure;

[0039] The target prediction result includes at least one of whether there is an injury on the corpse in the target image, injury location information, first injury type information, and second injury type information.

[0040] Preferably,

[0041] The prediction method further comprises:

[0042] After outputting the target prediction result representing the injury condition on the corpse in the target image, the prediction method further includes:

[0043] Comparing the target prediction result with the preset standard information to obtain a comparison result;

[0044] The comparison result is used to evaluate the reliability of the target prediction result obtained by the cadaver injury result prediction model. According to a third aspect of the present disclosure, a training system for a cadaver injury result prediction model is provided, the training system comprising:

[0045] A data acquisition module, used to acquire several groups of sample data; wherein any group of the sample data includes a sample image and sample association identification information corresponding to a preset injury site on the corpse in the sample image;

[0046] A model training module is used to use each group of sample images as the input of a target preset model, and to use the corresponding sample association identification information as the output of the target preset model, so as to train and obtain the corpse injury result prediction model for predicting the corpse injury condition in any image.

[0047] Preferably, the data acquisition module includes a first acquisition unit and a first annotation processing unit;

[0048] Wherein, the first acquisition unit is used to acquire a number of the sample images;

[0049] The first annotation processing unit is used to perform annotation processing on the sample image to obtain a first sample damage result corresponding to the sample image; the first sample damage result includes at least one of whether there is damage, damage location information, and first damage type information;

[0050] The model training module is also used to use the sample image as the input of the first preset model, and use the sample injury result corresponding to the sample image as the output of the first preset model, so as to train and obtain the corpse injury result prediction model for predicting the corpse injury condition in any image;

[0051] or,

[0052] The data acquisition module includes a second acquisition unit and a second annotation processing unit;

[0053] Wherein, the second acquisition unit is used to acquire a number of the sample images;

[0054] The second annotation processing unit is used to perform annotation processing on the sample image to obtain a second sample damage result corresponding to the sample image;

[0055] The model training module is also used to use the sample image as the input of the second preset model, and use the second sample injury result corresponding to the sample image as the output of the second preset model, so as to train and obtain the corpse injury result prediction model for predicting the injury condition corresponding to the corpse injury position in any image;

[0056] Wherein, the first injury type information is used to characterize the acceleration and deceleration injury type of the injury on the corpse;

[0057] The second injury type information is used to characterize the condition type of the injury on the corpse.

[0058] Preferably, the first preset model includes a DeepLabv3+ model;

[0059] and / or,

[0060] The second preset model includes a ResNet18-ASPP model;

[0061] The ResNet18-ASPP model includes a first convolutional layer, a first residual structure layer, a second residual structure layer, a third residual structure layer, a fourth residual structure layer, a perforated spatial pyramid pooling structure, and an output layer, which are connected in sequence.

[0062] Preferably, the training system further comprises a format conversion unit;

[0063] The format conversion unit is used to determine whether the sample image is a preset format image before the sample image is annotated, and if not, convert the sample image into a preset format image;

[0064] and / or,

[0065] The sample images include CT images, photographed images, single scanned images, or images obtained by superimposing multiple consecutive scanned images.

[0066] Preferably, the preset injury site includes the brain;

[0067] The injury type corresponding to the first injury type information includes accelerated craniocerebral injury and / or decelerated craniocerebral injury;

[0068] The injury type corresponding to the second injury type information includes at least one of cerebral contusion, subarachnoid hemorrhage, scalp hematoma, skull fracture, subdural hematoma, epidural hematoma and cerebral hemorrhage.

[0069] According to a fourth aspect of the present disclosure, a system for predicting cadaver injuries is provided, the prediction system comprising:

[0070] A target image acquisition module, used for acquiring a target image;

[0071] A result output module is used to input the target image into the cadaver injury result prediction model, and output the target prediction result representing the injury condition on the cadaver in the target image; wherein the cadaver injury result prediction model is obtained by using the training system as described in the third aspect of the present disclosure;

[0072] Wherein, the target prediction result includes at least one of whether there is damage on the body in the target image, damage location information, first damage type information, and second damage type information. Preferably, the prediction system further includes: a reliability determination unit;

[0073] Wherein, the reliability determination is used to compare the target prediction result with preset standard information to obtain a comparison result; wherein, the comparison result is used to evaluate the reliability of the target prediction result predicted by the corpse injury result prediction model. According to a fifth aspect of the present disclosure, an electronic device is provided, comprising a memory, a processor, and a computer program stored on the memory and used to run on the processor, and when the processor executes the computer program, it implements the training method of the corpse injury result prediction model described in the first aspect of the present disclosure, and / or the method for predicting corpse injuries described in the second aspect of the present disclosure.

[0074] According to a sixth aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the training method of the corpse injury result prediction model described in the first aspect of the present disclosure and / or the corpse injury prediction method described in the second aspect of the present disclosure are implemented.

[0075] According to the seventh aspect of the present disclosure, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the training method of the cadaver injury outcome prediction model described in the first aspect of the present disclosure, and / or the cadaver injury prediction method described in the second aspect of the present disclosure.

[0076] On the basis of being in accordance with the common sense in the art, the above-mentioned preferred conditions can be arbitrarily combined to obtain the preferred embodiments of the present disclosure.

[0077] The positive and progressive effects of this disclosure are:

[0078] Through the training method of the corpse injury result prediction model and the corpse injury prediction method provided in the present invention, the corpse injury results on the target image can be identified quickly and accurately, thereby improving the efficiency of the corpse injury result identification.

[0079] Furthermore, the present invention breaks the traditional mode in which forensic doctors rely solely on their personal experience to estimate the injury modes of acceleration and deceleration craniocerebral injuries, and uses artificial intelligence technology to improve the speed and accuracy of forensic craniocerebral injury identification; it uses big data to intelligently identify the injury mechanism of craniocerebral injuries, thereby reducing subjective and empirical errors; and the evidence obtained from imaging examinations is less destructive and highly shareable, which is not only easy to be accepted by the families of the deceased, reducing the work pressure and burden of forensic doctors, but also can provide intuitive images and data parameter forms, so that non-forensic professionals including judges, lawyers, etc. can also enjoy scientific and high-quality judicial appraisal services. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 A flowchart of a method for training a cadaver injury outcome prediction model provided by an exemplary embodiment 1 of the present disclosure;

[0081] Figure 2 This is a schematic diagram of a process for obtaining several groups of sample data in an exemplary embodiment 1 of the present disclosure;

[0082] Figure 3 This is another flow chart of obtaining several groups of sample data in an exemplary embodiment 1 of the present disclosure;

[0083] Figure 4 This is a schematic diagram of a process of outputting associated identification information in Embodiment 2 of the present disclosure;

[0084] Figure 5 This is a schematic diagram of the structure of cutting pages in Embodiment 2 of the present disclosure;

[0085] Figure 6 This is a schematic diagram of the structure after cutting in Example 2 of the present disclosure;

[0086] Figure 7Schematic diagram of the structure of the input image in Embodiment 2 of the present disclosure;

[0087] Figure 8 A schematic diagram of a module of a training system for a cadaver injury outcome prediction model in implementation 3 of the present disclosure;

[0088] Fig. 9 This is a schematic diagram of the modules of the system for predicting cadaver injuries in Embodiment 4 of the present disclosure;

[0089] Fig.10 It is a schematic diagram of the structure of the electronic device in Embodiment 5 of the present disclosure. DETAILED DESCRIPTION

[0090] The present disclosure is further described below by way of examples, but the present disclosure is not limited to the scope of the examples.

[0091] The prefixes such as "first" and "second" used in the embodiments of the present disclosure are only used to distinguish different description objects, and have no limiting effect on the position, order, priority, quantity or content of the objects being described. The use of prefixes such as ordinal numbers used to distinguish description objects in the embodiments of the present disclosure does not limit the objects being described. For the description of the objects being described, please refer to the context of the claims or embodiments.

[0092] The description of the present embodiment should not constitute an unnecessary restriction due to the use of such prefixes. In addition, in the description of the present embodiment, unless otherwise specified, the meaning of "multiple" is two or more.

[0093] In the embodiments of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0094] Example 1

[0095] Figure 1 A flowchart of a method for training a cadaver injury outcome prediction model provided by an exemplary embodiment of the present disclosure.

[0096] like Figure 1 As shown, this embodiment provides a method for training a cadaver injury result prediction model, and the training method includes:

[0097] S11: Acquire several groups of sample data; wherein any group of sample data includes a sample image and sample associated identification information of a preset injury site on a corpse in the corresponding sample image.

[0098] S12: Using each group of sample images as input of a target preset model, and using corresponding sample association identification information as output of the target preset model, so as to train a body injury result prediction model for predicting body injury conditions in any image.

[0099] Among them, in the embodiments of the present disclosure, the sample images include CT images, photographed images, single scanned images or images superimposed by multiple continuous scanned images. The sample images provided by the present disclosure are used to train the target preset graphics, thereby improving the accuracy of the corpse injury result prediction model.

[0100] The sample images in this embodiment are sample images obtained from a cadaver craniocerebral injury CT (computer tomography) image database. The sample images are manually annotated with associated identification information of preset injury sites on the corresponding images, and a deep learning model is selected according to the identified target. The associated identification information corresponding to the preset injury sites is imported into the deep learning model, thereby obtaining a cadaver injury result prediction model for predicting the injury condition of the cadaver in any image. The cadaver injury result prediction model is used to realize efficient, objective and accurate forensic craniocerebral injury identification injury result judgment, thereby improving the work efficiency of cadaver injury identification.

[0101] like Figure 2 As shown, in one implementation, the step of obtaining several groups of sample data includes:

[0102] S21: Acquire several sample images.

[0103] S22: performing labeling processing on the sample image to obtain a first sample damage result corresponding to the sample image.

[0104] The first sample damage result includes at least one of whether damage exists, damage location information, and first damage type information.

[0105] The preset injury site in this embodiment includes the brain.

[0106] In a specific implementation, the first injury type information is used to characterize the acceleration and deceleration injury type of the injury on the corpse; the injury type corresponding to the first injury type information includes acceleration craniocerebral injury and / or deceleration craniocerebral injury. The steps of using each group of sample images as the input of the target preset model and using the corresponding sample association identification information as the output of the target preset model to train a corpse injury result prediction model for predicting the corpse injury condition in any image include:

[0107] The sample image is used as the input of the first preset model, and the sample injury result corresponding to the sample image is used as the output of the first preset model, so as to train a corpse injury result prediction model for predicting the corpse injury condition in any image.

[0108] like Figure 3 As shown, in another implementation, the step of obtaining several groups of sample data includes:

[0109] S31: Acquire several sample images.

[0110] S32: performing labeling processing on the sample image to obtain a second sample damage result corresponding to the sample image.

[0111] The second sample damage result includes at least one of whether damage exists, damage location information, first damage type information, and second damage type information.

[0112] In a specific implementation, the second injury type information is used to characterize the type of injury on the corpse; the injury type corresponding to the second injury type information includes at least one of cerebral contusion, subarachnoid hemorrhage, scalp hematoma, skull fracture, subdural hematoma, epidural hematoma and cerebral hemorrhage.

[0113] The steps of using each group of sample images as the input of the preset model and using the corresponding sample association identification information as the output of the preset model to train a body injury result prediction model for predicting the body injury condition in any image include:

[0114] The sample image is used as the input of the second preset model, and the second sample injury result corresponding to the sample image is used as the output of the second preset model to train a corpse injury result prediction model for predicting the injury condition corresponding to the corpse injury position in any image.

[0115] In the disclosed embodiment, the sample image is annotated. The specific processing process is as follows: a sample image generally contains 10-30 slice sequences. Three consecutive slices in the CT sequence are selected as R, G, and B channels to synthesize a color image as a sample for model training, thereby simulating the process of forensic experts judging the type of injury. When making a judgment, in addition to observing a single slice, the forensic doctor will also make a comprehensive judgment based on the upper and lower slices.

[0116] The sample images are annotated as follows: if at least one of the three slices has characteristic injuries of acceleration injury, the sample is marked as acceleration injury; if at least one of the three slices has characteristic injuries of deceleration injury, the sample is marked as deceleration injury; the remaining samples are marked as normal.

[0117] In addition, in this embodiment, the sample image is annotated, specifically, multiple experts read the same sample image, and the results recognized by most experts are used as the gold standard. For example: investigate the injury process of the person being identified, question witnesses, view video materials, investigate the crime scene, etc., and clarify the results of deceleration or acceleration craniocerebral injury according to the actual investigation. In a specific implementation, when the sample image is annotated to obtain the corresponding injury location information, the specific method is as follows: rotate the sample image: execute according to the probability of 0.8, the maximum left rotation angle is 2, and the maximum right rotation angle is 2; swap the sample image left and right: execute according to the probability of 0.5; cut the random area of ​​the sample image: execute according to the probability of 0.5, the maximum degree of left cutting is 3, and the maximum degree of right cutting is 3; elastically distort the sample image: execute according to the probability of 0.75, the grid height is 3, the grid width is 3, and the amplitude is 6; randomly adjust the brightness of the sample image: execute according to the probability of 0.2, the minimum brightness factor is 0.85, and the maximum brightness factor is 1.15, so as to realize the operation of cutting the corresponding sample image.

[0118] In this embodiment, the first preset model includes the DeepLabv3+ model; and / or, the second preset model includes the ResNet18-ASPP model.

[0119] The ResNet18-ASPP model includes a first convolutional layer, a first residual structure layer, a second residual structure layer, a third residual structure layer, a fourth residual structure layer, a perforated spatial pyramid pooling structure, and an output layer, which are connected in sequence.

[0120] Among them, the atrous spatial pyramid pooling structure is the ASPP structure. The present disclosure first passes the input sample image through a convolution layer to extract coarse-grained image features, then passes through four residual structure modules to extract rich semantic features, and then passes through the ASPP module to further extract fine-grained semantic understanding capabilities, and finally passes through a linear layer to obtain the final classification result; that is, the present disclosure improves the semantic understanding ability of multiple slices of craniocerebral injury in the corpse injury result prediction model by using the ASPP structure, imitates the reading habits of forensic experts, and thereby improves the acceleration and deceleration classification accuracy of the model.

[0121] The second preset model in this embodiment captures multi-scale contextual information by applying convolution operations in parallel on multiple convolution kernels with different expansion rates, thereby effectively processing targets of different sizes and improving the model's adaptability to complex scenes (such as CT images).

[0122] In this embodiment, before labeling the sample image, the training method further includes:

[0123] It is determined whether the sample image is an image in a preset format; if not, the sample image is converted into an image in a preset format.

[0124] The sample image in this embodiment is a medical format image, such as DICOM format (a picture format); therefore, before the sample image is annotated, the training method further includes: converting the medical format image into a preset format image.

[0125] Among them, the preset format image can be an image in PNG format (a lossless compressed bitmap image format), an image in JPG format (an image file format), etc., to facilitate manual annotation; further, the present disclosure can realize the conversion of sample image formats through RadiAnt (a format conversion tool).

[0126] Since the classification criteria of the training set and validation set of the corpse injury result prediction model in this implementation are derived from the actual results of acceleration and deceleration injuries in real-life scenarios and do not involve expert judgment results, it is more accurate and reliable to predict the injury condition of the corpse using the corpse injury result prediction model in the disclosed embodiment.

[0127] The following examples specifically illustrate the implementation principle of the training method of the cadaver injury result prediction model disclosed in the present invention:

[0128] First, manual annotation was performed based on CT images of craniocerebral injuries in cadavers.

[0129] Specifically, researchers with professional knowledge and experience converted the DICOM format of CT images into PNG format through RadiAnt (a format conversion tool), and used Labelme (an image annotation tool) to manually identify and annotate various types of craniocerebral injuries. The annotation content included the location, range and label color of the injury. The label color setting defaulted to the color corresponding to different numbers in Labelme. For example: 1 for brain contusion, 2 for subarachnoid hemorrhage, 2 for scalp hematoma, 4 for skull fracture, 5 for subdural hematoma, 6 for epidural hematoma and 7 for cerebral hemorrhage, etc.

[0130] Secondly, the model is selected and the parameters are adjusted.

[0131] Select a deep learning model, such as the DeepLabv3+ model, which is composed of DeepLabv3+ and a decoding module. Its main functions are to accurately locate the target while ensuring resolution and to increase the model training speed while ensuring stability. The model is used to identify and segment craniocerebral injuries, judge accelerated craniocerebral injuries and decelerated craniocerebral injuries, and build artificial intelligence automatic recognition software; in the disclosed embodiment, the craniocerebral injury CT image segmentation model is trained by the DeepLabv3+ model; and the accelerated craniocerebral injury and decelerated craniocerebral injury classification model is trained by the DeepLabv3+ model to form an artificial intelligence recognition software.

[0132] At the same time, in this embodiment, the accuracy of the model is also verified through specific CT images.

[0133] For example, a single cadaver brain CT image in a forensic identification example is imported into artificial intelligence recognition software, and finally the classification results of the injury mode judgment for each case are output (such as the output of the recognition results of acceleration injury or deceleration injury of the brain CT image), the classification credibility, etc. are output; and the above-mentioned classification results, credibility, etc. are output in the form of reports.

[0134] In a specific implementation, the results of the artificial intelligence recognition software are output in the form of images and professional texts, and the software's evaluation value of the reliability of the recognition results is output. The recognition damage evaluation value is the Dice coefficient. Dice is a set similarity measurement indicator, which is usually used to calculate the similarity of two samples, and the value threshold is [0, 1]. It is often used for image segmentation in medical images. The best segmentation result is 1, and the worst result is 0. The classification evaluation value is the probability of the degree of credibility, which is unreliable at 0-0.2, generally reliable at 0.2-0.6, and reliable at 0.6-1 for user reference.

[0135] The present invention establishes artificial intelligence recognition software based on CT images, and uses the artificial intelligence recognition software to artificial intelligence recognize whether the craniocerebral injury of the target image is an accelerated craniocerebral injury or a decelerated craniocerebral injury. The specific recognition steps include:

[0136] (1) Input of images to be inspected:

[0137] Access the web page through a mobile phone, tablet or computer, import the images of acceleration and deceleration craniocerebral injuries of corpses obtained in forensic identification cases into the pre-established forensic craniocerebral injury intelligent segmentation software, and establish a connection with the network server;

[0138] (2) Intelligent calibration of craniocerebral injury images:

[0139] Select model 1 (i.e. DeepLabv3+) in the software, input a single CT image, or a whole set of multiple CT images; use DeepLabv3+ to segment the brain injury area of ​​each CT image, and then fuse the region of interest to the segmented input image, simulating the process of forensic experts focusing on the injury area.

[0140] Then, all the fused images of the CT tomographic images are superimposed in the order of slices to obtain a channel feature map covering all craniocerebral injury information. Finally, the Resnet18-ASPP network is used as the classifier of the craniocerebral injury classification system, and the seven categories of scalp hematoma, skull fracture, subdural hematoma, epidural hematoma, brain contusion, subarachnoid hemorrhage and cerebral hemorrhage are marked with corresponding colors, and the numbers and colors of the manual labels are compared, and no injury parts are marked.

[0141] At this point, the segmentation results of the craniocerebral injury sites in the image are output and displayed after background calculations, and the results are saved in the cloud and can be retrieved, called out, or downloaded repeatedly.

[0142] (3) Classification and discrimination of the mode of craniocerebral injury:

[0143] Based on the segmentation results obtained by DeepLabv3+, the data was further imported into Resnet18-ASPP; Resnet18-ASPP analyzed the correlation between the seven injury categories identified by DeepLabv3+, including scalp hematoma, skull fracture, subdural hematoma, epidural hematoma, brain contusion, subarachnoid hemorrhage and cerebral hemorrhage. After analysis and processing in the background, it classified the victim's craniocerebral injury CT scan into acceleration injury or deceleration injury, and finally output a complete analysis report.

[0144] The present invention breaks the traditional mode that forensic doctors rely on their personal experience to estimate the injury modes of acceleration and deceleration craniocerebral injuries. It uses artificial intelligence technology to improve the speed and accuracy of forensic craniocerebral injury identification. It uses big data to intelligently identify the injury mechanism of craniocerebral injury to reduce subjective and empirical errors. Moreover, the evidence obtained from imaging examinations is less destructive and highly shareable, which is not only easy to be accepted by the family members of the deceased, reducing the work pressure and burden of forensic doctors, but also can provide intuitive images and data parameter forms, so that non-forensic professionals including judges, lawyers, etc. can also enjoy scientific and high-quality judicial appraisal services.

[0145] Example 2

[0146] The present embodiment provides a method for predicting corpse injuries, the prediction method comprising: acquiring a target image, inputting the target image into a corpse injury result prediction model as in Embodiment 1, and outputting target association identification information of a target injury site on the corpse corresponding to the target image.

[0147] like Figure 4 As shown, in one embodiment, the step of outputting the associated identification information of the injured part of the corpse on the corresponding target image includes:

[0148] S41: Acquire a target image.

[0149] S42: Input the target image into the corpse injury result prediction model, and output the target prediction result representing the injury condition on the corpse in the target image.

[0150] Among them, the corpse injury result prediction model is obtained by using the training method as in Example 1; the target prediction result includes at least one of whether there is injury on the corpse in the target image, injury location information, first injury type information, and second injury type information.

[0151] In this embodiment, the prediction method further includes:

[0152] After obtaining the target prediction result representing the injury condition on the corpse in the output target image, the prediction method further includes:

[0153] The target prediction result is compared with the preset standard information to obtain the comparison result.

[0154] The comparison result is used to evaluate the reliability of the target prediction result obtained by the prediction model of the cadaver injury result. The prediction method in this embodiment also includes: obtaining a target sample image in the sample image library whose similarity with the target image meets a preset condition; obtaining sample association identification information corresponding to the target sample image; calculating the similarity between the sample association identification information and the target association identification information; and obtaining the reliability corresponding to the target association identification information based on the similarity analysis.

[0155] The following examples specifically illustrate the implementation principle of the method for predicting cadaver injuries disclosed herein:

[0156] like Figures 5 to 7 As shown, a device for automatically updating menus of a mobile phone comprises:

[0157] Access the web version of the automatic segmentation software for craniocerebral injury through mobile phones, tablets and computers;

[0158] Select the DeepLabv3+ model to perform image segmentation;

[0159] Input the craniocerebral injury CT images stored in the device;

[0160] Output segmentation results, which are saved in the cloud and can be downloaded repeatedly;

[0161] The segmentation results were input into the ResNet18-ASPP model to classify accelerated craniocerebral injury and decelerated craniocerebral injury.

[0162] The classification result report of accelerated craniocerebral injury and deceleration craniocerebral injury is output. The classification result report is permanently saved in the database cloud and users can download it repeatedly.

[0163] Through the method for predicting corpse injuries provided by the present disclosure, the corpse injury results on the target image can be identified quickly and accurately, thereby improving the efficiency of identifying corpse injury results.

[0164] Example 3

[0165] Corresponding to the aforementioned embodiment of the training method of the cadaver injury outcome prediction model, the present disclosure also provides an embodiment of the training system of the cadaver injury outcome prediction model.

[0166] Figure 8 A module schematic diagram of a training system for a cadaver injury outcome prediction model provided by an exemplary embodiment of the present disclosure, the system comprising:

[0167] The data acquisition module 100 is used to acquire several groups of sample data; wherein any group of sample data includes a sample image and sample association identification information corresponding to a preset injury site on a corpse in the sample image;

[0168] The model training module 200 is used to use each group of sample images as the input of the target preset model, and the corresponding sample association identification information as the output of the target preset model, so as to train a corpse injury result prediction model for predicting the corpse injury condition in any image.

[0169] The data acquisition module in this embodiment includes a first acquisition unit and a first annotation processing unit;

[0170] Wherein, the first acquisition unit is used to acquire a number of sample images;

[0171] The first annotation processing unit is used to perform annotation processing on the sample image to obtain a first sample damage result corresponding to the sample image; the first sample damage result includes at least one of whether there is damage, damage location information, and first damage type information;

[0172] The model training module is also used to use the sample image as the input of the first preset model, and use the sample injury result corresponding to the sample image as the output of the first preset model, so as to train and obtain a corpse injury result prediction model for predicting the corpse injury condition in any image;

[0173] or,

[0174] The data acquisition module includes a second acquisition unit and a second annotation processing unit;

[0175] Wherein, the second acquisition unit is used to acquire a number of sample images;

[0176] The second annotation processing unit is used to perform annotation processing on the sample image to obtain a second sample damage result corresponding to the sample image;

[0177] The model training module is also used to use the sample image as the input of the second preset model, and use the second sample injury result corresponding to the sample image as the output of the second preset model, so as to train and obtain a body injury result prediction model for predicting the injury condition corresponding to the body injury position in any image;

[0178] The first injury type information is used to characterize the acceleration and deceleration injury type of the injury on the corpse;

[0179] The second injury type information is used to characterize the condition type of the injury on the corpse.

[0180] The first preset model in this embodiment includes the DeepLabv3+ model;

[0181] and / or,

[0182] The second preset model includes the ResNet18-ASPP model;

[0183] The ResNet18-ASPP model includes a first convolutional layer, a first residual structure layer, a second residual structure layer, a third residual structure layer, a fourth residual structure layer, a perforated spatial pyramid pooling structure, and an output layer, which are connected in sequence.

[0184] The training system in this embodiment also includes a format conversion unit;

[0185] The format conversion unit is used to determine whether the sample image is a preset format image before labeling the sample image, and if not, convert the sample image into a preset format image;

[0186] and / or,

[0187] The sample images include CT images, photographed images, single scanned images, or images obtained by superimposing multiple consecutive scanned images.

[0188] The preset injury sites in this embodiment include the cranial brain;

[0189] The injury type corresponding to the first injury type information includes accelerated craniocerebral injury and / or decelerated craniocerebral injury;

[0190] The injury type corresponding to the second injury type information includes at least one of cerebral contusion, subarachnoid hemorrhage, scalp hematoma, skull fracture, subdural hematoma, epidural hematoma and cerebral hemorrhage.

[0191] As for the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The system embodiment described above is only illustrative, in which the units described as separate components may or may not be physically separated, and the components as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution.

[0192] The present invention breaks the traditional mode that forensic doctors rely on their personal experience to estimate the injury modes of acceleration and deceleration craniocerebral injuries. It uses artificial intelligence technology to improve the speed and accuracy of forensic craniocerebral injury identification. It uses big data to intelligently identify the injury mechanism of craniocerebral injury to reduce subjective and empirical errors. Moreover, the evidence obtained from imaging examinations is less destructive and highly shareable, which is not only easy to be accepted by the family members of the deceased, reducing the work pressure and burden of forensic doctors, but also can provide intuitive images and data parameter forms, so that non-forensic professionals including judges, lawyers, etc. can also enjoy scientific and high-quality judicial appraisal services.

[0193] Example 4

[0194] Corresponding to the aforementioned embodiments of the method for predicting cadaver injuries, the present disclosure also provides embodiments of the method for predicting cadaver injuries.

[0195] Fig. 9 A module schematic diagram of a system for predicting cadaver injuries provided by an exemplary embodiment of the present disclosure, the system comprising:

[0196] The target image acquisition module 300 is used to acquire a target image.

[0197] A result output module 400 is used to input the target image into the cadaver injury result prediction model, and output the target prediction result representing the injury condition on the cadaver in the target image; wherein the cadaver injury result prediction model is obtained by using the training system as in Example 3;

[0198] The target prediction result includes at least one of whether there is damage on the body in the target image, damage location information, first damage type information, and second damage type information. The prediction system in this embodiment further includes: a reliability determination unit;

[0199] Among them, the reliability determination is used to compare the target prediction result with the preset standard information to obtain the comparison result; wherein, the comparison result is used to evaluate the reliability of the target prediction result predicted by the corpse injury result prediction model. For the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can be referred to the partial description of the method embodiment. The system embodiment described above is merely schematic, wherein the units described as separate components may or may not be physically separated, and the components as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the disclosed scheme.

[0200] Through the prediction system for corpse injuries provided by the present disclosure, the corpse injury results on the target image can be identified quickly and accurately, thereby improving the efficiency of the identification of corpse injury results.

[0201] Example 5

[0202] Fig.10 This is a schematic diagram of the structure of an electronic device provided in Embodiment 5 of the present disclosure. The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the method in the above embodiment is implemented when the processor executes the program. Fig.10 The electronic device 30 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0203] like Fig.10 As shown, the electronic device 30 may be in the form of a general-purpose computing device, for example, it may be a server device. The components of the electronic device 30 may include, but are not limited to: at least one processor 31, at least one memory 32, and a bus 33 connecting different system components (including the memory 32 and the processor 31).

[0204] The bus 33 includes a data bus, an address bus, and a control bus.

[0205] The memory 32 may include a volatile memory, such as a random access memory (RAM) 321 and / or a cache memory 322 , and may further include a read-only memory (ROM) 323 .

[0206] The memory 32 may also include a program / utility 325 having a set (at least one) of program modules 324, such program modules 324 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0207] The processor 31 executes various functional applications and data processing by running the computer programs stored in the memory 32, such as the method in the above-mentioned embodiment 1 or 2 of the present disclosure.

[0208] The electronic device 30 may also communicate with one or more external devices 34 (e.g., keyboards, pointing devices, etc.). Such communication may be performed via an input / output (I / O) interface 35. Furthermore, the model generating device 30 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 36. Fig.10 As shown, the network adapter 36 communicates with other modules of the model-generated device 30 via the bus 33. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the model-generated device 30, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems, etc.

[0209] It should be noted that although several units / modules or sub-units / modules of the electronic device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules to be embodied.

[0210] Example 6

[0211] The embodiments of the present disclosure further provide a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method provided in any one of the embodiments 1 or 2 above.

[0212] The readable storage medium may include but is not limited to: a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device or any suitable combination of the above.

[0213] Example 7

[0214] The embodiments of the present disclosure also provide a computer program product, including a computer program, which implements the method provided in the above-mentioned embodiment 1 or 2 when executed by a processor.

[0215] Among them, the program code for executing the computer program product of the present disclosure can be written in any combination of one or more programming languages, and the program code can be executed completely on the user device, partially on the user device, as an independent software package, partially on the user device and partially on a remote device, or completely on the remote device.

[0216] Although the specific embodiments of the present disclosure are described above, those skilled in the art should understand that this is only an example, and the protection scope of the present disclosure is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present disclosure, but these changes and modifications all fall within the protection scope of the present disclosure.

Claims

1. A method for training a cadaver injury outcome prediction model, characterized in that: The training method comprises: Acquire several groups of sample data; wherein any group of the sample data includes a sample image and sample association identification information corresponding to a preset injury site on the corpse in the sample image; Using each group of the sample images as input of a target preset model, and using the corresponding sample association identification information as output of the target preset model, so as to train and obtain the body injury result prediction model for predicting the body injury condition in any image; The step of obtaining several groups of sample data comprises: Acquire a number of the sample images; Performing labeling processing on the sample image to obtain a first sample damage result corresponding to the sample image; The first sample damage result includes at least one of whether there is damage, damage location information, and first damage type information; The step of using each group of the sample images as the input of the target preset model and using the corresponding sample association identification information as the output of the target preset model to train the body injury result prediction model for predicting the body injury condition in any image includes: Using the sample image as input of a first preset model, and using the sample injury result corresponding to the sample image as output of the first preset model, so as to train and obtain the corpse injury result prediction model for predicting corpse injury conditions in any image; or, The step of obtaining several groups of sample data comprises: Acquire a number of the sample images; Performing labeling processing on the sample image to obtain a second sample damage result corresponding to the sample image; The second sample damage result includes at least one of whether there is damage, damage location information, first damage type information, and second damage type information; The step of using each group of sample images as the input of a preset model, and using the corresponding sample association identification information as the output of the preset model to train and obtain the corpse injury result prediction model for predicting corpse injury conditions in any image comprises: Using the sample image as input of a second preset model, and using a second sample injury result corresponding to the sample image as output of the second preset model, so as to train and obtain the corpse injury result prediction model for predicting the injury condition corresponding to the corpse injury position in any image; Wherein, the first injury type information is used to characterize the acceleration and deceleration injury type of the injury on the corpse; The second injury type information is used to characterize the condition type of the injury on the corpse.

2. The method for training a cadaver injury outcome prediction model according to claim 1, characterized in that: The first preset model includes a DeepLabv3+ model; and / or, The second preset model includes a ResNet18-ASPP model; The ResNet18-ASPP model includes a first convolutional layer, a first residual structure layer, a second residual structure layer, a third residual structure layer, a fourth residual structure layer, a perforated spatial pyramid pooling structure, and an output layer, which are connected in sequence.

3. The training method of the cadaver injury outcome prediction model according to claim 1 or 2, characterized in that: Before the sample image is annotated, the training method further includes: Determining whether the sample image is an image in a preset format, and if not, converting the sample image into an image in a preset format; and / or, The sample images include CT images, photographed images, single scanned images, or images obtained by superimposing multiple consecutive scanned images.

4. The method for training a cadaver injury outcome prediction model according to claim 1, characterized in that: The preset injury site includes the brain; The injury type corresponding to the first injury type information includes accelerated craniocerebral injury and / or decelerated craniocerebral injury; The injury type corresponding to the second injury type information includes at least one of cerebral contusion, subarachnoid hemorrhage, scalp hematoma, skull fracture, subdural hematoma, epidural hematoma and cerebral hemorrhage.

5. A method for predicting cadaver injuries, characterized in that: The prediction method comprises: Get the target image; Inputting the target image into a cadaver injury result prediction model, and outputting a target prediction result representing the injury condition on the cadaver in the target image; Wherein, the cadaver injury result prediction model is obtained by using the training method described in any one of claims 1 to 4; The target prediction result includes at least one of whether there is an injury on the corpse in the target image, injury location information, first injury type information, and second injury type information.

6. The method for predicting cadaver injuries according to claim 5, characterized in that: The prediction method further comprises: After outputting the target prediction result representing the injury condition on the corpse in the target image, the prediction method further includes: Comparing the target prediction result with the preset standard information to obtain a comparison result; The comparison result is used to evaluate the reliability of the target prediction result obtained by the corpse injury result prediction model.

7. A training system for a cadaver injury outcome prediction model, characterized in that: The training system comprises: A data acquisition module, used to acquire several groups of sample data; wherein any group of the sample data includes a sample image and sample association identification information corresponding to a preset injury site on the corpse in the sample image; A model training module, used to use each group of the sample images as an input of a target preset model, and use the corresponding sample association identification information as an output of the target preset model, so as to train and obtain the corpse injury result prediction model for predicting corpse injury conditions in any image; The data acquisition module includes a first acquisition unit and a first annotation processing unit; Wherein, the first acquisition unit is used to acquire a number of the sample images; The first annotation processing unit is used to perform annotation processing on the sample image to obtain a first sample damage result corresponding to the sample image; the first sample damage result includes at least one of whether there is damage, damage location information, and first damage type information; The model training module is also used to use the sample image as the input of the first preset model, and use the sample injury result corresponding to the sample image as the output of the first preset model, so as to train and obtain the corpse injury result prediction model for predicting the corpse injury condition in any image; or, The data acquisition module includes a second acquisition unit and a second annotation processing unit; The second acquisition unit is used to acquire a plurality of the sample images; The second annotation processing unit is used to perform annotation processing on the sample image to obtain a second sample damage result corresponding to the sample image; the second sample damage result includes at least one of whether there is damage, damage location information, first damage type information, and second damage type information; The model training module is also used to use the sample image as the input of the second preset model, and use the second sample injury result corresponding to the sample image as the output of the second preset model, so as to train and obtain the corpse injury result prediction model for predicting the injury condition corresponding to the corpse injury position in any image; Wherein, the first injury type information is used to characterize the acceleration and deceleration injury type of the injury on the corpse; The second injury type information is used to characterize the condition type of the injury on the corpse.

8. A system for predicting cadaver injuries, characterized in that: The prediction system comprises: A target image acquisition module, used for acquiring a target image; A result output module is used to input the target image into the cadaver injury result prediction model, and output the target prediction result representing the injury condition on the cadaver in the target image; wherein the cadaver injury result prediction model is obtained by using the training system according to claim 7; The target prediction result includes at least one of whether there is damage on the body in the target image, damage location information, first damage type information, and second damage type information.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and used to run on the processor, characterized in that: When the processor executes the computer program, it implements the training method of the cadaver injury result prediction model described in any one of claims 1 to 4, and / or the cadaver injury prediction method described in claim 5 or 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the training method of the cadaver injury result prediction model described in any one of claims 1 to 4, and / or the cadaver injury prediction method described in claim 5 or 6.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, it implements the training method of the cadaver injury result prediction model as described in any one of claims 1 to 4, and / or the cadaver injury prediction method as described in claim 5 or 6.

Images