TACE treatment evaluation method and training method and device of TACE treatment evaluation model

Through real-time segmentation and prediction of DSA image data, a TACE treatment evaluation model was constructed, which solved the problem of uncertainty in the treatment results of patients with middle and late stage HCC in TACE treatment and achieved real-time efficacy prediction and treatment optimization intraoperatively.

CN119069112APending Publication Date: 2024-12-03NEUSOFT MEDICAL SYST CO LTD
View PDF 0 Cites 4 Cited by

Patent Information

Application Number
CN202411092374.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-08
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

In the existing TACE treatment, the treatment results of patients with middle and advanced HCC vary from person to person, and TACE therapy may lead to worsening of liver function and affect prognosis. There is a lack of effective methods to estimate therapeutic responses to optimize treatment decisions.

Method used

By acquiring DSA image data, detecting liver areas, extracting the time difference data and liver characteristic data of keyframes, performing tumor segmentation and prediction, a TACE treatment evaluation model is constructed to achieve real-time automatic efficacy prediction intraoperatively and optimize treatment decisions.

Benefits of technology

Real-time segmentation and prediction of liver tumors during TACE treatment is achieved, providing accurate prediction results for treatment response, helping doctors optimize treatment plans and reduce the risk of liver function deterioration.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119069112A_ABST
    Figure CN119069112A_ABST
Patent Text Reader

Abstract

The invention discloses a TACE treatment evaluation method and a TACE treatment evaluation model training method and device.The TACE treatment evaluation method comprises the steps that DSA image data are acquired, and the DSA image data are image data including the liver acquired in the TACE treatment operation process; detecting the DSA image data to obtain a plurality of key frames including a liver region; performing feature extraction on the plurality of key frames to obtain time difference data and liver feature data of the plurality of key frames; performing tumor segmentation based on the key frame, the time difference data and the liver feature data to obtain a tumor segmentation result; and performing prediction based on the tumor segmentation result to obtain a TACE treatment prediction result. According to the TACE treatment evaluation method, real-time automatic curative effect prediction can be carried out on DSA image data in an operation, and a treatment decision is optimized according to a prediction result.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to technical fields such as medical treatment and TACE efficacy, and in particular to a TACE treatment evaluation method, a training method and device for a TACE treatment evaluation model. Background Art

[0002] Hepatocellular carcinoma (HCC) is the sixth most common malignant tumor in the world. According to international guidelines, transcatheter arterial chemoembolization (TACE) is currently the standard treatment for patients with advanced and unresectable HCC. However, the advanced stage of HCC is determined based on liver function and tumor burden, and the patient's condition varies from person to person, which leads to different treatment outcomes, with only 15-55% of patients responding to treatment. In addition, some patients may experience worsening liver function after TACE. If the patient's liver function further deteriorates due to repeated TACE cycles, this may have a negative impact on prognosis and may hinder subsequent anti-tumor treatment. Therefore, a predictive model for estimating TACE treatment response is crucial as a reference for clinical decision-making. Summary of the invention

[0003] To this end, the purpose of the embodiments of the present application is to propose a TACE treatment evaluation method, device, training method, device, electronic device, storage medium and computer program product of a TACE treatment evaluation model, which can perform real-time automatic efficacy prediction of DSA image data during surgery and optimize treatment decisions based on the prediction results.

[0004] An embodiment of the present application provides a TACE treatment evaluation method, which includes: acquiring DSA image data, wherein the DSA image data is image data including the liver acquired during the TACE treatment operation; detecting the DSA image data to obtain multiple key frames including the liver area; performing feature extraction on the multiple key frames to obtain time difference data and liver feature data of the multiple key frames; performing tumor segmentation based on the key frames, the time difference data and the liver feature data to obtain a tumor segmentation result; and performing prediction based on the tumor segmentation result to obtain a TACE treatment prediction result.

[0005] Exemplarily, the extracting features from the multiple key frames to obtain time difference data and liver feature data of the multiple key frames includes: extracting features from the multiple key frames through a tumor segmentation model to obtain time difference data and liver feature data of the multiple key frames.

[0006] Exemplarily, the tumor segmentation model includes a timing difference molecular model and a liver region segmentation sub-model, and the feature extraction of the multiple key frames through the tumor segmentation model to obtain time difference data and liver feature data of the multiple key frames includes: extracting features of the multiple key frames based on the timing difference molecular model to obtain time difference data of the multiple key frames; and extracting features of at least one of the multiple key frames based on the liver region segmentation sub-model to obtain liver feature data.

[0007] Exemplarily, the tumor segmentation model also includes a final fusion segmentation sub-model, and the tumor segmentation is performed based on the key frames, the time difference data and the liver feature data to obtain a tumor segmentation result, including: using the final fusion segmentation sub-model, based on the time difference data and the liver feature data, performing tumor identification on the key frames, and performing tumor segmentation on the key frames based on the tumor identification results to obtain a tumor segmentation result.

[0008] Exemplarily, the predicting based on the tumor segmentation result to obtain a TACE treatment prediction result includes: predicting based on the tumor segmentation result and clinical data associated with TACE treatment through a therapeutic efficacy prediction model to obtain a TACE treatment prediction result.

[0009] Exemplarily, the efficacy prediction model includes an image feature extraction module and a multi-layer perception module. The efficacy prediction model is used to make a prediction based on the tumor segmentation result and the clinical data associated with TACE treatment to obtain a TACE treatment prediction result, including: performing feature extraction on the clinical data based on the multi-layer perception module to obtain clinical data features; performing feature extraction on the tumor segmentation result based on the image feature extraction module to obtain tumor region features; and performing a prediction based on the clinical data features and the tumor region features to obtain a TACE treatment prediction result.

[0010] Exemplarily, the efficacy prediction model also includes a fully connected layer and a normalization layer, and the prediction based on the clinical data features and the tumor region features to obtain a TACE treatment prediction result includes: inputting the clinical data features and the tumor region features into the fully connected layer, and inputting the output result of the fully connected layer into the normalization layer to obtain the TACE treatment prediction result, wherein the output of the fully connected layer is connected to the input of the normalization layer, and the TACE treatment prediction result includes the probability of effectiveness of the TACE treatment.

[0011] Another embodiment of the present application provides a training method for a TACE treatment evaluation model, wherein the TACE treatment evaluation model includes at least a tumor segmentation model and a therapeutic effect prediction model, and the method includes: acquiring multiple key sample frames including a liver area, tumor outer contour reference data, and follow-up data characterizing TACE treatment results, wherein the multiple key samples are obtained by detecting DSA image sample data; performing feature extraction on the multiple key sample frames based on the tumor segmentation model to be trained to obtain time difference data and liver feature data of the multiple key sample frames, performing tumor segmentation based on the key sample frames, the time difference data, and the liver feature data to obtain a tumor segmentation result, training the tumor segmentation model to be trained based on the tumor segmentation result and the tumor outer contour reference data to obtain a trained tumor segmentation model; predicting the tumor segmentation result based on the therapeutic effect prediction model to be trained to obtain a TACE treatment prediction result, and training the therapeutic effect prediction model to be trained based on the TACE treatment prediction result and the follow-up data to obtain a trained tumor segmentation model.

[0012] Exemplarily, the tumor segmentation model to be trained is trained based on the tumor segmentation result and the tumor outer contour reference data to obtain a trained tumor segmentation model, including: determining the loss function value of the tumor segmentation model based on a pre-constructed loss function, the tumor segmentation result and the tumor outer contour reference data; training the tumor segmentation model to be trained based on the loss function value to obtain a trained tumor segmentation model.

[0013] Exemplarily, the loss function value includes a first loss function value, a second loss function value, a third loss function value and a fourth loss function value, and the loss function value of the tumor segmentation model is determined based on a pre-constructed loss function, the tumor segmentation result and the tumor outer contour reference data, including: determining the first loss function value of the timing difference sub-model, the second loss function value of the liver region segmentation sub-model and the third loss function value of the final fusion segmentation sub-model based on the pre-constructed loss function, the tumor segmentation result and the tumor outer contour reference data; determining the fourth loss function value of the tumor segmentation model based on the first loss function value, the second loss function value and the third loss function value, wherein the fourth loss function value is obtained by weighting the first loss function value, the second loss function value and the third loss function value.

[0014] Another embodiment of the present application provides a TACE treatment evaluation device, which includes: an acquisition module, used to acquire DSA image data, wherein the DSA image data is image data including the liver acquired during the TACE treatment operation; a detection module, used to detect the DSA image data to obtain multiple key frames including the liver area; an extraction module, used to extract features from the multiple key frames to obtain time difference data and liver feature data of the multiple key frames; a segmentation module, used to perform tumor segmentation based on the key frames, the time difference data and the liver feature data to obtain a tumor segmentation result; and a prediction module, used to perform prediction based on the tumor segmentation result to obtain a TACE treatment prediction result.

[0015] Another embodiment of the present application provides a training device for a TACE treatment evaluation model, wherein the TACE treatment evaluation model includes at least a tumor segmentation model and a therapeutic effect prediction model, and the device includes: an acquisition module, used to acquire multiple key sample frames including a liver area, tumor outer contour reference data and follow-up data characterizing TACE treatment results, wherein the multiple key samples are obtained by detecting DSA image sample data; a first training module, used to perform feature extraction on the multiple key sample frames based on the tumor segmentation model to be trained, obtain time difference data and liver feature data of the multiple key sample frames, perform tumor segmentation based on the key sample frames, the time difference data and the liver feature data to obtain a tumor segmentation result, train the tumor segmentation model to be trained based on the tumor segmentation result and the tumor outer contour reference data to obtain a trained tumor segmentation model; a second training module, used to predict the tumor segmentation result based on the therapeutic effect prediction model to be trained, obtain a TACE treatment prediction result, and train the therapeutic effect prediction model to be trained based on the TACE treatment prediction result and the follow-up data to obtain a trained tumor segmentation model.

[0016] Another embodiment of the present application provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method of any of the above embodiments when executing the computer program.

[0017] Another embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method of any of the above embodiments are implemented.

[0018] Another embodiment of the present application provides a computer program product, which includes instructions. When the instructions are executed by a processor of a computer device, the computer device is enabled to perform the steps of the method of any of the above embodiments.

[0019] In the above implementation, DSA image data is obtained, wherein the DSA image data is image data including the liver obtained during the TACE treatment operation; the DSA image data is detected to obtain multiple key frames including the liver area; feature extraction is performed on the multiple key frames to obtain time difference data and liver feature data of the multiple key frames; tumor segmentation is performed based on the key frames, time difference data and liver feature data to obtain tumor segmentation results; prediction is performed based on the tumor segmentation results to obtain TACE treatment prediction results. The TACE treatment evaluation method of the present invention can perform real-time automatic efficacy prediction on DSA image data during surgery and optimize treatment decisions based on the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A flowchart of a TACE treatment evaluation method provided for an embodiment of the present application;

[0021] Figure 2 A schematic diagram of a TACE treatment evaluation method provided in an embodiment of the present application;

[0022] Figure 3 A schematic diagram of a target detection model provided for an embodiment of the present application;

[0023] Figure 4 An architectural diagram of a tumor segmentation model provided in an embodiment of the present application;

[0024] Figure 5 A flowchart for obtaining TACE treatment prediction results provided in an embodiment of the present application;

[0025] Figure 6 A schematic diagram of a therapeutic efficacy prediction model provided for an embodiment of the present application;

[0026] Figure 7 A flowchart of a training method for a TACE treatment evaluation model provided in an embodiment of the present application;

[0027] Figure 8 A flowchart of obtaining a trained tumor segmentation model provided in an embodiment of the present application;

[0028] Fig. 9 A flowchart for determining the loss function value of a tumor segmentation model provided in an embodiment of the present application;

[0029] Fig.10 A schematic diagram of a TACE treatment evaluation device provided in an embodiment of the present application;

[0030] Fig.11 A schematic diagram of a training device for a TACE treatment evaluation model provided in an embodiment of the present application;

[0031] Fig.12 A block diagram of an electronic device provided for an embodiment of the present application. DETAILED DESCRIPTION

[0032] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0033] Hepatocellular carcinoma (HCC) is the sixth most common malignant tumor in the world. According to international guidelines, transcatheter arterial chemoembolization (TACE) is currently the standard treatment for patients with advanced and unresectable HCC. However, the advanced stage of HCC is determined based on liver function and tumor burden, and the patient's condition varies from person to person, which leads to different treatment outcomes, with only 15-55% of patients responding to treatment. In addition, some patients may experience worsening liver function after TACE. If the patient's liver function further deteriorates due to repeated TACE cycles, this may have a negative impact on prognosis and may hinder subsequent anti-tumor treatment. Therefore, a predictive model for estimating TACE treatment response is crucial as a reference for clinical decision-making.

[0034] Based on this, the present application proposes a TACE treatment evaluation method, which can predict the TACE efficacy of DSA image data in real time during surgery to help doctors optimize treatment decisions based on the prediction results.

[0035] Figure 1 is a flow chart of a TACE treatment evaluation method according to an embodiment of the present invention.

[0036] As an example, Figure 1 As shown, TACE treatment evaluation methods include:

[0037] S101, acquiring DSA image data, wherein the DSA image data is image data including the liver acquired during a TACE treatment operation.

[0038] S102, detecting the DSA image data to obtain a plurality of key frames including the liver region.

[0039] S103, extracting features from the multiple key frames to obtain time difference data and liver feature data of the multiple key frames.

[0040] S104, performing tumor segmentation based on the key frame, the time difference data and the liver feature data to obtain a tumor segmentation result.

[0041] S105, performing prediction based on the tumor segmentation result to obtain a TACE treatment prediction result.

[0042] Exemplarily, digital subtraction angiography (DSA) is an important technology in the TACE treatment process. DSA image data can provide real-time and dynamic information such as lesion location, catheter navigation, and arterial blood supply. At present, the prediction of TACE efficacy is based on ultrasound angiography (US), magnetic resonance imaging (MRI), or computed tomography (CT) images before treatment, but the efficacy of TACE mainly depends on real-time arterial blood supply and tumor load. The above imaging modalities cannot provide relevant information about the efficacy of TACE during surgery, and angiography can be used to directly observe this information during TACE. Therefore, this application predicts the efficacy of TACE based on DSA image data.

[0043] Exemplarily, firstly, DSA image data including the liver during the TACE treatment operation is acquired, the DSA image data is detected, and multiple key frames including the liver region are obtained, and then the multiple key frames are feature extracted to obtain the time difference data and liver feature data of the multiple key frames. Tumor segmentation is performed based on the key frames, time difference data and liver feature data to obtain the tumor segmentation result. Prediction is performed based on the tumor segmentation result to obtain the TACE treatment prediction result. It can be understood that the liver region is first detected on the DSA image data to locate the approximate region of interest in order to better segment the tumor region, and then multiple key frames are obtained by cropping, and then the tumor region in the region of interest is continued to be segmented to obtain the tumor segmentation result, and finally the tumor segmentation result is predicted to obtain the TACE treatment prediction result.

[0044] The TACE treatment evaluation method of the present application can perform real-time automatic efficacy prediction on DSA image data during surgery and optimize treatment decisions based on the prediction results.

[0045] Figure 2 It is an architectural diagram of a TACE treatment evaluation method according to an embodiment of the present application.

[0046] like Figure 2As shown, the TACE treatment evaluation model includes a liver detection model, a tumor segmentation model and a efficacy prediction model. The DSA video is continuously analyzed during the operation. The liver detection model is first used to detect and obtain multiple key frames of the liver area. The tumor segmentation model is then used to segment the tumor area in the multiple key frames to obtain the tumor segmentation result. The tumor segmentation result is input into the efficacy prediction model to obtain the TACE treatment prediction result, wherein the TACE treatment prediction result can be the effective probability of TACE efficacy.

[0047] As an example, detecting the DSA image data to obtain a plurality of key frames including the liver region includes: detecting the DSA image data using a target detection model to obtain a plurality of key frames including the liver region.

[0048] For example, the DSA image data is first detected using the target detection model to obtain a plurality of key frames including the liver region. Figure 3 The target detection model diagram shown in Figure 1 is

[0049] For example, Figure 3 The schematic diagram of the target detection model shown in the figure is used to determine the approximate ROI (Region of Interest) of the liver. The DSA angiography image is input, and then the feature map of the image is extracted through a set of convolution, ReLU and pooling layers (conv+relu+pooling) of the basic network. These feature maps are shared for the subsequent region proposal network layer (RPN) and fully connected layer. The region proposal network is used to generate the proposed region. This layer uses softmax to determine whether the anchors belong to positive samples (positive) or negative samples (negative), and then uses bounding box regression (bounding box regression) to correct the anchors to obtain accurate proposed regions. It should be noted that the classification (softmax) and positioning (Bbox regression) in the region proposal network are carried out in parallel. The key is that they share the same features and predict at the same time, rather than in sequence. Then, combined with the input feature map and the proposed region information, it is sent to the subsequent fully connected layer after passing through the ROI pooling layer (ROIPooling) to determine the target category, and the final accurate position of the detection box is obtained again through bounding box regression. Finally, the bounding box of the liver region is labeled and output, and the results can be visualized or saved. During the forward propagation of the target detection network, classification and localization are performed simultaneously rather than sequentially. This design not only improves computational efficiency, but also utilizes the information complementarity brought by feature sharing to improve detection performance.

[0050] This application uses a target detection model to pre-detect tumor lesions, locate the tumor lesion area, and improve the accuracy of tumor segmentation. Before segmenting the tumor, the non-ROI area is reduced by locating the liver area to ensure that the segmentation model focuses on the ROI area of ​​the tumor, thereby improving the accuracy of tumor segmentation.

[0051] The ROI area of ​​the tumor obtained by the target detection model is cropped to obtain multiple key frames for better segmentation of the tumor area.

[0052] As an example, feature extraction is performed on multiple key frames to obtain time difference data and liver feature data of multiple key frames, including: feature extraction is performed on multiple key frames through a tumor segmentation model to obtain time difference data and liver feature data of multiple key frames.

[0053] Exemplarily, the tumor segmentation model is used to segment the tumor area. Considering the particularity of DSA video having a certain time difference, this application adopts a liver temporal difference (LTD) submodule, a liver region segmentation (LRS) submodel and a final fusion segmentation (FFS) submodel to form a tumor segmentation model. Multiple key frames are input into the tumor segmentation model, and the tumor segmentation model extracts features from the multiple key frames to obtain time difference data and liver feature data of the multiple key frames.

[0054] As an example, the tumor segmentation model includes a timing difference molecular model and a liver region segmentation sub-model. The tumor segmentation model is used to extract features from multiple key frames to obtain time difference data and liver feature data for the multiple key frames, including: extracting features from multiple key frames based on the timing difference molecular model to obtain time difference data for the multiple key frames; and extracting features from at least one of the multiple key frames based on the liver region segmentation sub-model to obtain liver feature data.

[0055] For example, Figure 4 As shown, the temporal difference molecular model, the liver region segmentation sub-model and the final fusion segmentation sub-model can all use nnU-Net as the model backbone network. The temporal difference molecular model extracts features from multiple key frames to obtain time difference data of multiple key frames. The liver region segmentation sub-model extracts features from at least one of the multiple key frames to obtain liver feature data, which can be understood as the liver region mask data predicted by the liver region segmentation sub-model.

[0056] This application achieves more accurate liver and tumor area segmentation by fusing the dynamic information change characteristics of time difference. In order to learn the dynamic changes of multiple consecutive frames in digital subtraction angiography DSA video, the tumor segmentation model adopts temporal difference learning (LTD) technology. LTD uses continuous frames to capture how the tumor appearance changes over time, and subtracts the information of adjacent frames from the image information of a specific frame. Therefore, the model can effectively detect subtle changes in tumor characteristics, thereby obtaining more accurate tumor area segmentation results.

[0057] As an example, the tumor segmentation model also includes a final fusion segmentation sub-model, which performs tumor segmentation based on key frames, time difference data, and liver feature data to obtain tumor segmentation results, including:

[0058] Through the final fusion segmentation sub-model, tumor recognition is performed on key frames based on time difference data and liver feature data, and tumor segmentation is performed on key frames based on the tumor recognition results to obtain tumor segmentation results.

[0059] Exemplarily, the time difference data, liver feature data and key frames are input into the final fusion segmentation sub-model. The final fusion segmentation sub-model performs tumor identification on the key frames based on the time difference data and liver feature data to accurately obtain tumor area data, and performs tumor segmentation on the key frames based on the tumor identification results to obtain tumor segmentation results.

[0060] For example, Figure 4 As shown, the three inputs of the final fusion segmentation sub-model are the key frame of the detected liver area, the liver area mask predicted by LRS, and the time difference learned by LTD. The key frame, the time difference learned by LTD, and the liver area mask predicted by LRS are connected and input into the final fusion segmentation sub-model to finally obtain the tumor segmentation result.

[0061] As an example, based on the tumor segmentation results, the TACE treatment prediction results are obtained, including:

[0062] The efficacy prediction model is used to make predictions based on the tumor segmentation results and clinical data associated with TACE treatment to obtain TACE treatment prediction results.

[0063] Exemplarily, after segmenting and obtaining the tumor segmentation result, the tumor region is predicted, and the efficacy prediction model of the present application can be used to predict TACE treatment for the tumor region. Before predicting TACE treatment, clinical data associated with TACE treatment needs to be obtained, and the efficacy prediction model is used to predict the tumor segmentation result and the clinical data associated with TACE treatment to obtain the TACE treatment prediction result.

[0064] Among them, the clinical data include: gender, age, TACE date, history of hepatitis B, preoperative antiviral treatment, cirrhosis, hypertension, diabetes, ALP, ALT, AST, GGT, PT, TB, AFP, albumin, platelets, Child-Pugh grade, BCLC stage, and postoperative short-term efficacy evaluation. Among them, the objective response rate and overall response rate (ORR) can be selected as endpoints for the postoperative short-term efficacy evaluation.

[0065] As an example, Figure 5 As shown, the efficacy prediction model includes an image feature extraction module and a multi-layer perception module. Through the efficacy prediction model, predictions are made based on tumor segmentation results and clinical data associated with TACE treatment to obtain TACE treatment prediction results, including:

[0066] S501, extracting features of clinical data based on a multi-layer perception module to obtain clinical data features.

[0067] S502: extract features from the tumor segmentation result based on the image feature extraction module to obtain tumor region features.

[0068] S503, making predictions based on clinical data characteristics and tumor region characteristics to obtain TACE treatment prediction results.

[0069] For example, Figure 6 The schematic diagram of the efficacy prediction model shown in FIG. 1 includes an image feature extraction module and a multi-layer perception module. The tumor segmentation result obtained by the tumor segmentation model is input into the image feature extraction module. The image feature extraction module includes a 3D convolution layer, a 3D pooling layer, a residual block 1-4 and a 3D pooling layer connected in sequence to obtain tumor region features. The clinical data associated with TACE treatment collected during surgery is input into the multi-layer perception module. The multi-layer perception machine extracts features from the clinical data to obtain clinical data features. For example, the multi-layer perception machine can encode and represent the clinical data. Combining the clinical data features and the tumor region features for prediction, a TACE treatment prediction result is obtained.

[0070] As an example, the efficacy prediction model also includes a fully connected layer and a normalization layer, which performs prediction based on clinical data features and tumor area features to obtain TACE treatment prediction results, including: inputting clinical data features and tumor area features into the fully connected layer, and inputting the output result of the fully connected layer into the normalization layer to obtain the TACE treatment prediction result, wherein the output of the fully connected layer is connected to the input of the normalization layer, and the TACE treatment prediction result includes the probability of TACE treatment being effective.

[0071] For example, Figure 6As shown in FIG. 1 , the efficacy prediction model further includes a fully connected layer and a normalization layer. The outputs of the image feature extraction module and the multi-layer perception module are connected to the fully connected layer. The fully connected layer receives clinical data features and tumor region features. The fully connected layer is connected to the normalization layer, where the normalization layer can be a softmax function. The output of the fully connected layer is connected to the input of the normalization layer, and the normalization layer outputs the TACE treatment prediction result, for example, the TACE treatment effectiveness probability.

[0072] This application uses DSA video information to build a system for real-time automatic prediction of TACE treatment efficacy during surgery to help clinicians adjust treatment plans in a timely manner and provide clinical decision support. It also integrates image features with patient clinical data information to achieve more accurate prediction of treatment response.

[0073] This application also proposes a training method for a TACE treatment evaluation model.

[0074] As an example, Figure 7 As shown, the training method of the TACE treatment evaluation model includes:

[0075] S701, acquiring multiple key sample frames including a liver region, tumor outer contour reference data, and follow-up data representing TACE treatment results, wherein the multiple key samples are obtained by detecting DSA image sample data.

[0076] S702, based on the tumor segmentation model to be trained, feature extraction is performed on multiple key sample frames to obtain time difference data and liver feature data of the multiple key sample frames, tumor segmentation is performed based on the key sample frames, time difference data and liver feature data to obtain a tumor segmentation result, and the tumor segmentation model to be trained is trained based on the tumor segmentation result and tumor outer contour reference data to obtain a trained tumor segmentation model.

[0077] S703, predicting the tumor segmentation result based on the efficacy prediction model to be trained to obtain a TACE treatment prediction result, and training the efficacy prediction model to be trained based on the TACE treatment prediction result and follow-up data to obtain a trained tumor segmentation model.

[0078] Exemplarily, before training the TACE treatment assessment model, training data is first collected.

[0079] Data collection includes: DSA imaging data, clinical information of patients, and follow-up information. Follow-up information, for example, can be used to determine whether the tumor has become smaller after three months, or whether there is still tumor in the pathological section. Follow-up information is used as reference data to measure the accuracy of the prediction model. The follow-up results after TACE treatment are used as the gold standard for efficacy. For example, the gold standard for cancer patients to be cured is the absence of tumor in the pathological section.

[0080] Inclusion criteria: (1) HCC (hepatocellular carcinoma) confirmed by imaging or pathology; (2) first time receiving TACE (transarterial chemoembolization) treatment; (3) liver CT imaging within 1 month before treatment and 2 months after treatment; (4) no previous anti-tumor treatment.

[0081] Exclusion criteria: (1) patients over 18 years old; (2) patients who had received other anti-tumor treatments in the past; (3) patients with other tumors or extrahepatic metastases; (4) patients with missing information or lost to follow-up.

[0082] The personnel were initially screened according to the inclusion criteria, and the exclusion criteria were screened again based on the inclusion criteria.

[0083] Analyze imaging features, including: tumor location, tumor number, tumor size, tumor enhancement degree, whether the tumor edge is clear, whether there is bleeding / necrosis in the tumor, whether there is an artery in the tumor, whether there is portal vein cancer thrombus, whether there are satellite lesions, etc.

[0084] Exemplarily, the collected data is preprocessed, and the preprocessing includes:

[0085] (1) At least two experienced doctors outline the outer contour of the lesion region (ROI) respectively, and repeat the outline again after a period of time to retain the corresponding lesion region segmentation (Mask) information. This is used in the tumor segmentation model to train the segmentation model for segmentation accuracy.

[0086] Among them, 3D-Slicer software can be used to outline the lesion area. CT takes the arterial phase sequence, and DSA takes 6 frames of images selected in equal proportion from the beginning of contrast agent injection to the complete staining of the tumor in the video for model training.

[0087] (2) Use the z-score method to standardize the image to ensure the stability of image features.

[0088] (3) Repeatability testing was performed on the features extracted from the ROIs drawn by different doctors at different times, and the intra-group correlation coefficient was calculated. That is, different doctors drew the ROIs separately, and a “group” represented the outer contours of the same lesion area drawn by different doctors. The consistency of the two results was then compared. The higher the coefficient, the better the consistency.

[0089] As an example, the TACE treatment evaluation model also includes a target detection model to obtain multiple key sample frames including the liver area, including: detecting the DSA image data through the target detection model to obtain multiple key sample frames including the liver area.

[0090] Exemplarily, after data processing, the target detection model detects the DSA image data to obtain a plurality of key sample frames including the liver region.

[0091] Then, based on the tumor segmentation model to be trained, feature extraction is performed on multiple key sample frames to obtain time difference data and liver feature data of multiple key sample frames, tumor segmentation is performed based on the key sample frames, time difference data and liver feature data to obtain tumor segmentation results, and the tumor segmentation model to be trained is trained based on the tumor segmentation results and tumor outer contour reference data to obtain a trained tumor segmentation model.

[0092] Then, the tumor segmentation result is predicted based on the efficacy prediction model to be trained to obtain the TACE treatment prediction result, and the efficacy prediction model to be trained is trained based on the TACE treatment prediction result and follow-up data to obtain a trained tumor segmentation model.

[0093] The target detection model, the trained tumor segmentation model and the trained tumor segmentation model are combined to obtain the required TACE treatment evaluation model.

[0094] As an example, the tumor segmentation model to be trained includes a timing difference molecular model and a liver region segmentation sub-model. Feature extraction is performed on multiple key sample frames based on the tumor segmentation model to be trained to obtain time difference data and liver feature data of multiple key sample frames, including: feature extraction is performed on multiple key sample frames based on the timing difference molecular model to obtain time difference data of multiple key sample frames; feature extraction is performed on at least one of the key sample frames based on the liver region segmentation sub-model to obtain liver feature data.

[0095] As an example, the tumor segmentation model also includes a final fusion segmentation sub-model, which performs tumor segmentation based on key sample frames, time difference data and liver feature data to obtain tumor segmentation results, including: using the final fusion segmentation sub-model, based on the time difference data and liver feature data, performing tumor identification on key frames, and performing tumor segmentation on key frames based on the tumor identification results to obtain tumor segmentation results.

[0096] As an example, Figure 8 As shown, the tumor segmentation model to be trained is trained based on the tumor segmentation result and the tumor outer contour reference data to obtain a trained tumor segmentation model, including:

[0097] S801, determining a loss function value of a tumor segmentation model based on a pre-constructed loss function, a tumor segmentation result, and tumor outer contour reference data.

[0098] S802, training the tumor segmentation model to be trained based on the loss function value to obtain a trained tumor segmentation model.

[0099] Exemplarily, the output result of the tumor segmentation model to be trained is the tumor segmentation result, and the tumor outer contour reference data may be the tumor outer contour reference data collected in advance and drawn by a doctor. The segmentation accuracy of the tumor segmentation model is judged based on the tumor outer contour reference data.

[0100] Exemplarily, a pre-constructed loss function is used to determine the loss function value of the tumor segmentation model based on the tumor segmentation results and the reference data of the tumor outer contour. The tumor segmentation model to be trained is trained with the goal of minimizing the loss function value of the tumor segmentation model to obtain a trained tumor segmentation model.

[0101] As an example, Fig. 9 As shown, based on the pre-constructed loss function, the tumor segmentation result and the reference data of the tumor outer contour, the loss function value of the tumor segmentation model is determined, including:

[0102] S901, based on the pre-constructed loss function, tumor segmentation results and tumor outer contour reference data, determine the first loss function value of the time difference sub-model, the second loss function value of the liver area segmentation sub-model and the third loss function value of the final fusion segmentation sub-model.

[0103] S902, based on the first loss function value, the second loss function value and the third loss function value, determine the fourth loss function value of the tumor segmentation model, wherein the fourth loss function value is obtained by weighting the first loss function value, the second loss function value and the third loss function value.

[0104] Exemplarily, the loss function value includes a first loss function value, a second loss function value, a third loss function value, and a fourth loss function value. Because the tumor segmentation model includes a timing difference sub-model, a liver region segmentation sub-model, and a final fusion segmentation sub-model, correspondingly, each sub-model corresponds to a loss function value. The first loss function value corresponds to the timing difference sub-model, the second loss function value corresponds to the liver region segmentation sub-model, the third loss function value corresponds to the final fusion segmentation sub-model, and the fourth loss function value corresponds to the tumor segmentation model. The calculation formulas for the first loss function value, the second loss function value, the third loss function value, and the fourth loss function value are as follows:

[0105] L LTD =|I LTD -I FD | L1 (1)

[0106] L LRS =a*|I LRS -I LM | BCE +(1-a)*|I LRS -I LM | DICE (2)

[0107] L seg =a*|I seg -I GT | BCE +(1-a)*|I seg -I GT | DICE (3)

[0108] L Total =λ0*L LTD +λ1*L LRS +L seg (4)

[0109] Among them, L LTD represents the loss function value of the time series difference molecular model, L LRS represents the loss function value of the liver segmentation sub-model, Lseg represents the loss function value of the final fusion segmentation sub-model, L Total It is the total loss function value of the tumor segmentation model representing the entire network, that is, the fourth loss function value. I represents the image of each frame, a, λO and λ1 are hyperparameters used to control the effect of the loss function. FD represents the frame difference, LM represents the liver region mask, and GT represents the tumor mask value. Among them, the fourth loss function value is obtained by weighting the first loss function value, the second loss function value and the third loss function value.

[0110] Exemplarily, the learning temporal difference sub-model and the liver region segmentation sub-model and the final fusion segmentation sub-model are simultaneously trained according to the loss function value. The tumor segmentation model uses five-fold cross validation on the training cohort, is optimized on the validation cohort, and is evaluated on the test cohort.

[0111] This application mines the difference information between frames, allowing the tumor segmentation model to understand the temporal changes in tumor characteristics, helping to highlight the changes in tumor staining, shape and size, and improving the overall segmentation accuracy for real-time evaluation of treatment response.

[0112] As an example, predicting the tumor segmentation result based on the efficacy prediction model to be trained to obtain the TACE treatment prediction result includes: predicting the tumor segmentation result and clinical data associated with TACE treatment based on the efficacy prediction model to be trained to obtain the TACE treatment prediction result.

[0113] As an example, the efficacy prediction model to be trained includes an image feature extraction module and a multi-layer perception module. The tumor segmentation result is predicted based on the efficacy prediction model to be trained to obtain a TACE treatment prediction result, including: extracting features of clinical data based on the multi-layer perception module to obtain clinical data features; extracting features of tumor segmentation results based on the image feature extraction module to obtain tumor region features; and predicting based on clinical data features and tumor region features to obtain a TACE treatment prediction result.

[0114] For example, the trained efficacy prediction model includes an image feature extraction module and a multi-layer perception module. During the training process, the features of clinical information and image information are learned simultaneously to improve the prediction performance. The network structure used by the prediction model is as follows: Figure 6 As shown, the specific network parameters are: the parameters of the 3D convolution layer can be 7×7×7,64,stride(1,2,2), the parameters of the 3D pooling layer are 3×3×3,max pool,stride(2,2,2), and the parameters of the residual block 1 are The parameters of residual block 2 are The parameters of residual block 3 are The parameters of residual block 4 can be, The Adam optimizer was selected for the efficacy prediction model, the learning rate was set to 0.001, the batch size was 16, and 30 epochs were trained in the supervised learning phase. The follow-up results were used as the basis for judging the efficacy prediction model to measure its prediction accuracy.

[0115] As an example, the efficacy prediction model also includes a fully connected layer and a normalization layer, which performs prediction based on clinical data features and tumor area features to obtain TACE treatment prediction results, including: inputting clinical data features and tumor area features into the fully connected layer, and inputting the output result of the fully connected layer into the normalization layer to obtain the TACE treatment prediction result, wherein the output of the fully connected layer is connected to the input of the normalization layer, and the TACE treatment prediction result includes the probability of TACE treatment being effective.

[0116] The present application also proposes a TACE treatment evaluation device.

[0117] As an example, Fig.10As shown, the TACE treatment evaluation device includes: an acquisition module 1001, used to acquire DSA image data, wherein the DSA image data is image data including the liver acquired during the TACE treatment operation; a detection module 1002, used to detect the DSA image data to obtain multiple key frames including the liver area; an extraction module 1003, used to extract features from the multiple key frames to obtain time difference data and liver feature data of the multiple key frames; a segmentation module 1004, used to perform tumor segmentation based on the key frames, the time difference data and the liver feature data to obtain a tumor segmentation result; a prediction module 1005, used to perform prediction based on the tumor segmentation result to obtain a TACE treatment prediction result.

[0118] This application also proposes a training device for the ACE treatment evaluation model.

[0119] As an example, Fig.11 As shown, the training device of the TACE treatment evaluation model includes:

[0120] The acquisition module 1101 is used to acquire multiple key sample frames including the liver area, tumor outer contour reference data and follow-up data representing the TACE treatment results, wherein the multiple key samples are obtained by detecting the DSA image sample data; the first training module 1102 is used to extract features from the multiple key sample frames based on the tumor segmentation model to be trained, obtain time difference data and liver feature data of the multiple key sample frames, perform tumor segmentation based on the key sample frames, the time difference data and the liver feature data to obtain a tumor segmentation result, train the tumor segmentation model to be trained based on the tumor segmentation result and the tumor outer contour reference data to obtain a trained tumor segmentation model; the second training module 1103 is used to predict the tumor segmentation result based on the efficacy prediction model to be trained, obtain the TACE treatment prediction result, and train the efficacy prediction model to be trained based on the TACE treatment prediction result and the follow-up data to obtain a trained tumor segmentation model.

[0121] The application also proposes a computer-readable storage medium.

[0122] In this embodiment, a computer program is stored on a computer-readable storage medium, and when the computer program is executed by a processor, the steps of the above-mentioned TACE treatment evaluation method and the training method of the TACE treatment evaluation model are implemented.

[0123] Fig.12 A block diagram of an electronic device provided for an embodiment of the present application.

[0124] An embodiment of the present application provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the above-mentioned TACE treatment evaluation method and the training method of the TACE treatment evaluation model are implemented.

[0125] like Fig.12 As shown, for ease of understanding, the embodiment of the present application shows a specific electronic device.

[0126] Electronic devices are intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0127] like Fig.12 As shown, the device includes a computing unit 1201, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1202 or a computer program loaded from a storage unit 1208 into a random access memory (RAM) 1203. In RAM 1203, various programs and data required for the operation of the electronic device can also be stored. The computing unit 1201, ROM 1202, and RAM 1203 are connected to each other via a bus 1204. An input / output (I / O) interface 1205 is also connected to the bus 1204.

[0128] Multiple components in the electronic device are connected to the I / O interface 1205, including: an input unit 1206, such as a keyboard, a mouse, etc.; an output unit 1207, such as various types of displays, speakers, etc.; a storage unit 1208, such as a disk, an optical disk, etc.; and a communication unit 1209, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 1209 allows the electronic device to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0129] The computing unit 1201 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1201 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 1201 executes the various methods described above, such as the TACE treatment evaluation method and the training method of the TACE treatment evaluation model. For example, in some embodiments, the TACE treatment evaluation method and the training method of the TACE treatment evaluation model may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 1208. In some embodiments, part or all of the computer program may be loaded and / or installed on an electronic device via ROM 1202 and / or a communication unit 1209. When the computer program is loaded into RAM 1203 and executed by the computing unit 1201, the TACE treatment evaluation method and the training method of the TACE treatment evaluation model described above may be executed. Alternatively, in other embodiments, the computing unit 1201 may be configured to execute the TACE treatment evaluation method and the training method of the TACE treatment evaluation model in any other appropriate manner (eg, by means of firmware).

[0130] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purposes of this application, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection portion with one or more wirings (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or otherwise processing in a suitable manner if necessary, and then stored in a computer memory.

[0131] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0132] In the description of the present application, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In the present application, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0133] In the description of the present application, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present application.

[0134] In addition, the terms "first", "second", etc. used in the embodiments of the present application are only used for descriptive purposes and should not be understood as indicating or implying relative importance, or implicitly indicating the number of technical features indicated in the present embodiment. Therefore, the features defined by the terms "first", "second", etc. in the embodiments of the present application can explicitly or implicitly indicate that at least one of the features is included in the embodiment. In the description of the present application, the word "multiple" means at least two or two or more, such as two, three, four, etc., unless otherwise clearly and specifically defined in the embodiments.

[0135] In this application, unless otherwise clearly specified or limited in the embodiments, the terms "installed", "connected", "connected" and "fixed" etc. appearing in the embodiments should be understood in a broad sense. For example, the connection can be a fixed connection, a detachable connection, or an integrated connection. It can be understood that it can also be a mechanical connection, an electrical connection, etc.; of course, it can also be a direct connection, or an indirect connection through an intermediate medium, or it can be the internal connection of two elements, or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to the specific implementation situation.

[0136] In the present application, unless otherwise clearly specified and limited, a first feature being “above” or “below” a second feature may mean that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, a first feature being “above”, “above”, and “above” a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. A first feature being “below”, “below”, and “below” a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is lower in level than the second feature.

[0137] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A TACE treatment evaluation method, It is characterized in that The method comprises: Acquiring DSA image data, wherein the DSA image data is image data including the liver acquired during a TACE treatment operation; Detecting the DSA image data to obtain a plurality of key frames including a liver region; Extracting features from the multiple key frames to obtain time difference data and liver feature data of the multiple key frames; Perform tumor segmentation based on the key frame, the time difference data and the liver feature data to obtain a tumor segmentation result; A prediction is performed based on the tumor segmentation result to obtain a TACE treatment prediction result.

2. The TACE treatment evaluation method according to claim 1, It is characterized in that The step of extracting features from the multiple key frames to obtain time difference data and liver feature data of the multiple key frames includes: The multiple key frames are subjected to feature extraction through a tumor segmentation model to obtain time difference data of the multiple key frames and liver feature data.

3. The TACE treatment evaluation method according to claim 2, It is characterized in that The tumor segmentation model includes a time difference sub-model and a liver region segmentation sub-model. The feature extraction of the multiple key frames by the tumor segmentation model to obtain the time difference data of the multiple key frames and the liver feature data includes: Extracting features of the multiple key frames based on the time difference submodel to obtain time difference data of the multiple key frames; Based on the liver region segmentation sub-model, feature extraction is performed on at least one of the multiple key frames to obtain liver feature data.

4. The TACE treatment evaluation method according to claim 2, It is characterized in that The tumor segmentation model also includes a final fusion segmentation sub-model, and the tumor segmentation is performed based on the key frame, the time difference data and the liver feature data to obtain a tumor segmentation result, including: Through the final fusion segmentation sub-model, tumor recognition is performed on key frames based on the time difference data and the liver feature data, and tumor segmentation is performed on the key frames based on the tumor recognition results to obtain tumor segmentation results.

5. The TACE treatment evaluation method according to claim 1, It is characterized in that The step of performing prediction based on the tumor segmentation result to obtain a TACE treatment prediction result includes: The efficacy prediction model is used to make a prediction based on the tumor segmentation result and clinical data associated with TACE treatment to obtain a TACE treatment prediction result.

6. The TACE treatment evaluation method according to claim 5, It is characterized in that The efficacy prediction model includes an image feature extraction module and a multi-layer perception module. The efficacy prediction model is used to predict based on the tumor segmentation result and clinical data associated with TACE treatment to obtain a TACE treatment prediction result, including: Extracting features from the clinical data based on the multi-layer perception module to obtain clinical data features; Performing feature extraction on the tumor segmentation result based on the image feature extraction module to obtain tumor region features; Prediction is performed based on the clinical data characteristics and the tumor region characteristics to obtain a TACE treatment prediction result.

7. The TACE treatment evaluation method according to claim 6, It is characterized in that The efficacy prediction model further includes a fully connected layer and a normalization layer. The prediction based on the clinical data characteristics and the tumor region characteristics to obtain the TACE treatment prediction result includes: The clinical data features and the tumor area features are input into the fully connected layer, and the output result of the fully connected layer is input into the normalization layer to obtain a TACE treatment prediction result, wherein the output of the fully connected layer is connected to the input of the normalization layer, and the TACE treatment prediction result includes the TACE treatment effectiveness probability.

8. A training method for a TACE treatment evaluation model. It is characterized in that The TACE treatment evaluation model at least includes a tumor segmentation model and an efficacy prediction model, and the method includes: Acquire multiple key sample frames including liver regions, tumor outer contour reference data, and follow-up data characterizing TACE treatment results, wherein the multiple key samples are obtained by detecting DSA image sample data; Based on the tumor segmentation model to be trained, feature extraction is performed on the multiple key sample frames to obtain time difference data and liver feature data of the multiple key sample frames, tumor segmentation is performed based on the key sample frames, the time difference data and the liver feature data to obtain a tumor segmentation result, and the tumor segmentation model to be trained is trained based on the tumor segmentation result and tumor outer contour reference data to obtain a trained tumor segmentation model; The tumor segmentation result is predicted based on the efficacy prediction model to be trained to obtain a TACE treatment prediction result, and the efficacy prediction model to be trained is trained based on the TACE treatment prediction result and the follow-up data to obtain a trained tumor segmentation model.

9. The training method of the TACE treatment evaluation model according to claim 8, It is characterized in that The step of training the tumor segmentation model to be trained based on the tumor segmentation result and the tumor outer contour reference data to obtain a trained tumor segmentation model includes: Determining a loss function value of a tumor segmentation model based on a pre-constructed loss function, the tumor segmentation result, and the tumor outer contour reference data; The tumor segmentation model to be trained is trained based on the loss function value to obtain a trained tumor segmentation model.

10. According to the training method of the TACE treatment evaluation model according to claim 9, It is characterized in that The loss function value includes a first loss function value, a second loss function value, a third loss function value and a fourth loss function value, and the loss function value of the tumor segmentation model is determined based on the pre-constructed loss function, the tumor segmentation result and the tumor outer contour reference data, including: Based on the pre-constructed loss function, the tumor segmentation result and the tumor outer contour reference data, determine a first loss function value of the time difference sub-model, a second loss function value of the liver region segmentation sub-model and a third loss function value of the final fusion segmentation sub-model; Based on the first loss function value, the second loss function value and the third loss function value, a fourth loss function value of the tumor segmentation model is determined, wherein the fourth loss function value is obtained by weighting the first loss function value, the second loss function value and the third loss function value.

11. A TACE treatment evaluation device, It is characterized in that The device comprises: An acquisition module, used for acquiring DSA image data, wherein the DSA image data is image data including the liver acquired during the TACE treatment operation; A detection module, used for detecting the DSA image data to obtain a plurality of key frames including a liver region; An extraction module, used for performing feature extraction on the multiple key frames to obtain time difference data and liver feature data of the multiple key frames; A segmentation module, used for performing tumor segmentation based on the key frame, the time difference data and the liver feature data to obtain a tumor segmentation result; The prediction module is used to make predictions based on the tumor segmentation results to obtain TACE treatment prediction results.

12. A training device for a TACE treatment evaluation model, It is characterized in that The TACE treatment evaluation model at least includes a tumor segmentation model and an efficacy prediction model, and the device includes: An acquisition module, used to acquire a plurality of key sample frames including a liver region, reference data of a tumor outer contour, and follow-up data representing TACE treatment results, wherein the plurality of key samples are obtained by detecting DSA image sample data; A first training module is used to extract features from the multiple key sample frames based on the tumor segmentation model to be trained, obtain time difference data and liver feature data of the multiple key sample frames, perform tumor segmentation based on the key sample frames, the time difference data and the liver feature data to obtain a tumor segmentation result, and train the tumor segmentation model to be trained based on the tumor segmentation result and tumor outer contour reference data to obtain a trained tumor segmentation model; The second training module is used to predict the tumor segmentation result based on the efficacy prediction model to be trained to obtain a TACE treatment prediction result, and to train the efficacy prediction model to be trained based on the TACE treatment prediction result and the follow-up data to obtain a trained tumor segmentation model.

13. An electronic device comprising a memory and a processor, wherein the memory stores a computer program. It is characterized in that When the processor executes the computer program, the steps of the method according to any one of claims 1 to 10 are implemented.

14. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 10 are implemented.

Citation Information

Cited By

  • Image prediction tumor treatment reactivity classification method, device, equipment and medium

    CN119693724A

  • Imaging Prediction of Tumor Treatment Responsiveness Classification Method, Device, Equipment and Medium

    CN119693724B

  • Method and product for training brain perfusion image sequence segmentation model

    CN120580243A

  • Method and product for training a model for segmenting a sequence of brain perfusion images

    CN120580243B