Liver cancer TACE postoperative active focus automatic detection and segmentation model and construction method thereof

Abstract

The invention discloses a liver cancer TACE postoperative active focus automatic detection and segmentation model and a construction method thereof, and the construction method comprises the steps: obtaining a TACE postoperative MR image of a liver cancer patient as a training sample, and dividing the training sample into a training set and a verification set; performing digital subtraction processing on the MR images in the training set to generate digital subtraction images corresponding to the advanced artery period, the portal vein period and the delay period; based on a deep learning network, performing deep learning network training by jointly training each stage image of the MR image in the set and the corresponding digital subtraction image to obtain a liver cancer TACE postoperative active focus automatic detection and segmentation model; and verifying the detection and segmentation efficiency of the model through a verification set. The model provided by the invention can realize automatic detection and segmentation of the liver cancer TACE postoperative active lesion, can automatically obtain the TACE postoperative active lesion information of the liver cancer patient, has high accuracy, and effectively assists a clinician in accurately diagnosing the treatment effect of the liver cancer patient after intervention.

Description

Technical Field

[0001] The present invention relates to an automatic detection and segmentation model for active foci after TACE surgery for liver cancer and a construction method thereof, and belongs to the technical field of molecular biology research and bioinformatics. Background Art

[0002] Primary liver cancer is the second most deadly tumor in my country, among which hepatocellular carcinoma (HCC, hereinafter referred to as liver cancer) is the most common type of primary liver cancer, accounting for 75%-85% of all primary liver cancer patients. Most HCCs are already in the middle and late stages when diagnosed, and transarterial chemoembolization (TACE) is the preferred treatment for mid-stage HCC, and is widely used in my country. However, due to the widespread heterogeneity of patients and tumor individuals, active lesions usually remain after TACE. At present, imaging evaluation after TACE, such as the mRECIST standard or EASL standard, focuses on the measurement of the size of enhanced lesions, or the LI-RADSTRA v2024 standard focuses on the diagnosis of active lesions. Due to the diverse manifestations of active lesions after TACE for liver cancer, clinical diagnosis is challenging. In the evaluation of the efficacy of liver cancer after TACE, the consistency of the evaluation of the efficacy of liver cancer after TACE is low due to inconsistent clinical experience and excessive reliance on the subjective judgment of doctors. Therefore, automatic detection and segmentation of active lesions after TACE for liver cancer is of great clinical significance for improving clinical diagnosis efficiency, enhancing the accuracy of active lesion evaluation, and guiding clinicians in making subsequent treatment decisions for patients. Summary of the invention

[0003] The purpose of the present invention is to provide a model for automatic detection and segmentation of active foci after TACE for liver cancer and a method for constructing the same. The present invention analyzes multi-sequence information of follow-up MR images of active foci after TACE for liver cancer, uses a deep learning network combined with an image processing method of digital subtraction, and constructs a model for automatic detection and segmentation of active foci after TACE for liver cancer, so as to achieve rapid and accurate diagnosis of active foci after TACE for liver cancer, thereby solving the gaps in previous research on active foci and the problems of low efficiency, accuracy and consistency in postoperative diagnosis of active foci.

[0004] To achieve the above object, the present invention adopts the following technical solution:

[0005] A method for constructing an automatic detection and segmentation model for active lesions after TACE surgery for liver cancer comprises the following steps:

[0006] Step S1, obtaining MR images of liver cancer patients after TACE as training samples, and dividing them into a training set and a validation set; the MR images at least include plain scan phase, late arterial phase, portal venous phase and delayed phase images;

[0007] Step S2, performing digital subtraction processing on the MR images in the training set obtained in step S1 to generate digital subtraction images corresponding to the late arterial phase, portal venous phase and delayed phase;

[0008] Step S3, based on the deep learning network, each phase of the MR images in the training set and their corresponding digital subtraction images are combined to perform deep learning network training to obtain an automatic detection and segmentation model for active lesions after TACE for liver cancer;

[0009] Step S4, verifying the detection and segmentation performance of the model through a validation set.

[0010] Furthermore, in step S1, the MR image is a follow-up enhanced MR image of a patient who underwent TACE for liver cancer.

[0011] Furthermore, in step S1, the training samples include labeling information of active lesions after TACE surgery for liver cancer.

[0012] Furthermore, the marking information of active lesions after TACE for liver cancer is obtained by the following steps: first, the MR images are evaluated based on LR-TRAv2024, and the lesions with an evaluation result of LR-TR viable are included as research objects, i.e., active lesions. Then, the ROI of the active lesions is outlined on the MR images in the training samples, and the ROI is outlined in the plain scan phase, late arterial phase, portal venous phase, and delayed phase of the MR images.

[0013] Furthermore, in step S1, the training samples containing the labeling information of active lesions after TACE surgery for liver cancer are divided into a training set and a validation set, and subsequent steps S2 and S3 are performed based on the training set and the validation set.

[0014] Furthermore, the training samples are randomly divided into a training set and a validation set.

[0015] Furthermore, in step S2, different phases of the MR image are registered, and digital subtraction processing is performed using the enhanced late arterial phase, portal venous phase, delayed phase and plain scan phase of the MR image to obtain a digital subtraction image.

[0016] An automatic detection and segmentation model for active lesions after TACE surgery for liver cancer is obtained through the construction method, and can input MR images of liver cancer patients after TACE surgery and output volume information of active lesions after TACE surgery for liver cancer.

[0017] Beneficial effects: The present invention uses the enhanced MR images of the follow-up after TACE for liver cancer, and constructs a model for automatic detection and segmentation of active foci after TACE for liver cancer based on a deep learning network combined with digital subtraction images. Compared with existing research, the present invention fills the gap in automatic detection and segmentation of active foci after TACE, and the model has good credibility and reliability. Through the model, the diagnostic speed and accuracy of clinicians in diagnosing active foci after TACE for liver cancer are improved, providing a reference for clinicians to formulate individualized treatment plans for liver cancer patients after TACE, and solving the problem of low efficiency in clinical diagnosis of active foci after TACE for liver cancer. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a diagnostic performance based on the application of adaptive deep learning network nn-Unet combined with digital subtraction images in the automatic detection of active lesions in HCC patients after TACE.

[0019] Figure 2 It is a comparison of the volume of active foci automatically segmented by the automatic detection and segmentation model based on the deep learning network and the volume of manually annotated active foci; the ICC value of the volume of active foci between the automatic detection and segmentation system and the manually annotated one is 0.98, 95% CI: 0.98-0.99. Figure 3 This is the output result of the model used in the follow-up enhanced MR image of a patient with liver cancer after TACE in the embodiment. DETAILED DESCRIPTION

[0020] The present invention provides a method for constructing an automatic detection and segmentation model of active lesions after TACE surgery for liver cancer, comprising the following steps:

[0021] (1) Obtaining MR images of patients with liver cancer after TACE as training samples. Preferably, the MR images are follow-up enhanced MR images of patients who underwent TACE for liver cancer after surgery; the MR images at least include plain scan phase, late arterial phase, portal venous phase, and delayed phase images;

[0022] After acquiring MR images, the MR images were evaluated based on LR-TRAv2024, and the lesions with the evaluation results of LR-TR viable were included as research objects, i.e., active lesions. Then, the ROI of the active lesions was delineated on the MR images in the training samples. The ROI was delineated in the plain scan phase, late arterial phase, portal venous phase, and delayed phase of the MR images. A training data set containing the labeling information of active lesions after TACE for liver cancer was obtained, and it was randomly divided into a training set and a validation set.

[0023] (2) performing digital subtraction processing on the MR images in the training set obtained in step (1) to generate digital subtraction images corresponding to the late arterial phase, portal venous phase, and delayed phase;

[0024] The steps of digital subtraction processing are: registering different phases of the MR image, performing digital subtraction processing using the enhanced late arterial phase, portal venous phase, delayed phase and plain scan phase of the MR image to obtain a digital subtraction image;

[0025] (3) Based on the deep learning network, the images of each phase of MR images in the training set and their corresponding digital subtraction images were combined to train the deep learning network and obtain an automatic detection and segmentation model for active lesions after TACE for liver cancer;

[0026] (4) Verify the detection and segmentation performance of the model through a validation set.

[0027] An automatic detection and segmentation model for active lesions after TACE surgery for liver cancer is obtained through the construction method, and can input MR images of liver cancer patients after TACE surgery and output volume information of active lesions after TACE surgery for liver cancer.

[0028] The technical solution of the present invention is clearly and completely described below in conjunction with specific embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] Unless otherwise specified, the experimental methods used in the following examples are all conventional methods; the materials and reagents used are reagents and materials that can be obtained from commercial channels unless otherwise specified.

[0030] Example:

[0031] 1. Case selection

[0032] Patient inclusion criteria: clinical and imaging diagnosis of hepatocellular carcinoma; first-time TACE treatment; postoperative MRI plain scan + enhanced follow-up examination; postoperative follow-up MRI images were assessed as viable lesions according to LR-TRA (LI-RADS Treatment Response Algorithm) v2024.

[0033] 2. Data acquisition and analysis

[0034] Follow-up enhanced MR images of patients with liver cancer after TACE were collected, including at least the plain scan phase, late arterial phase, portal venous phase, and delayed phase. The enhanced MR images of patients with liver cancer after TACE were evaluated based on LR-TRAv2024. Specifically, the enhanced MR images of patients with liver cancer after TACE were evaluated for the efficacy of TACE treatment based on LR-TRAv2024, and the lesions evaluated as viable, i.e., active lesions, were screened. The ROI of the active lesions was manually outlined in each phase of the enhanced MR images (plain scan phase, late arterial phase, portal venous phase, delayed phase). All the outline results were randomly divided into a training set and a validation set at a ratio of 8:2. Based on the training set and the validation set, a deep learning network was combined with digital subtraction images for training and validation.

[0035] The different phases of the enhanced MR images of the follow-up after TACE for liver cancer were registered, and the enhanced phases (late arterial phase, portal venous phase, delayed phase) of the enhanced MR images were digitally subtracted from the plain scan phase to obtain the digital subtraction images.

[0036] 3. Construction of automatic detection and segmentation model for active lesions after TACE in liver cancer

[0037] The training set images were used using the adaptive deep learning network nn-Unet and combined with digital subtraction images to automatically detect and segment active lesions in patient images.

[0038] In this example, patients who were first diagnosed with hepatocellular carcinoma and received TACE treatment were retrospectively included. The inclusion criteria for patients were: clinical and imaging diagnosis of hepatocellular carcinoma; first TACE treatment; postoperative MRI plain scan + enhanced follow-up examination; postoperative follow-up MRI images were evaluated as viable lesions according to LR-TRA (LI-RADS Treatment Response Algorithm) v2024. A total of 260 cases of multicenter training set data and 65 cases of multicenter test set data were included.

[0039] In this embodiment, the automatic detection and segmentation model of active lesions after TACE in liver cancer has good diagnostic performance in a multi-center test set, and the Dice coefficient of the model for active lesion segmentation is 0.69 (95% CI: 0.54-0.78). Figure 1 The consistency between the volume of the active foci segmented by the model and the volume of the manually marked active foci is 0.98 (95% CI: 0.98-0.99). Figure 2 shown. Figure 3Shown is the output result of the model used in the follow-up enhanced MR image of a patient with liver cancer after TACE surgery in this embodiment. Figures A, C, and E show the same active lesion at different levels of the enhanced MR image, and Figures B, D, and F show the output results of the corresponding active lesion automatic detection and segmentation model, which have high accuracy.

[0040] In summary, the present invention innovatively aims at the automatic detection and segmentation of active lesions after TACE surgery for liver cancer, integrates multi-sequence active lesion imaging data, and constructs an automatic detection and segmentation model for active lesions after TACE surgery for hepatocellular carcinoma, providing a more reliable tool for the accurate diagnosis of active lesions after TACE surgery for clinical liver cancer.

[0041] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A method for constructing an automatic detection and segmentation model for active lesions after TACE surgery for liver cancer, characterized by: The following steps are involved: Step S1, obtaining MR images of liver cancer patients after TACE as training samples, and dividing them into a training set and a validation set; the MR images at least include plain scan phase, late arterial phase, portal venous phase and delayed phase images; Step S2, performing digital subtraction processing on the MR images in the training set obtained in step S1 to generate digital subtraction images corresponding to the late arterial phase, portal venous phase and delayed phase; Step S3, based on the deep learning network, each phase of the MR images in the training set and their corresponding digital subtraction images are combined to perform deep learning network training to obtain an automatic detection and segmentation model for active lesions after TACE for liver cancer; Step S4, verifying the detection and segmentation performance of the model through a validation set.

2. A method for constructing an automatic detection and segmentation model for active lesions after TACE surgery for liver cancer according to claim 1, characterized in that: In step S1, the MR image is a follow-up enhanced MR image of a patient who underwent TACE for liver cancer.

3. A method for constructing an automatic detection and segmentation model for active lesions after TACE surgery for liver cancer according to claim 1, characterized in that: In step S1, the training samples include labeling information of active lesions of liver cancer after TACE surgery.

4. A method for constructing an automatic detection and segmentation model for active lesions after TACE surgery for liver cancer according to claim 3, characterized in that: The labeling information of active lesions after TACE for liver cancer is obtained by the following steps: first, the MR images are evaluated based on LR-TRAv2024, and the lesions with an evaluation result of LR-TR viable are included as research objects, i.e., active lesions. Then, the ROI of the active lesions is delineated on the MR images in the training samples, and the ROI is delineated in the plain scan phase, late arterial phase, portal venous phase, and delayed phase of the MR images.

5. A method for constructing an automatic detection and segmentation model for active lesions after TACE surgery for liver cancer according to any one of claims 1 to 4, characterized in that: In the step S1, the training samples containing the labeling information of active lesions after TACE surgery for liver cancer are divided into a training set and a validation set, and the subsequent steps S2 and S3 are performed based on the training set and the validation set.

6. A method for constructing an automatic detection and segmentation model for active lesions after TACE surgery for liver cancer according to claim 5, characterized in that: The training samples are randomly divided into a training set and a validation set.

7. A method for constructing an automatic detection and segmentation model for active lesions after TACE surgery for liver cancer according to claim 1, characterized in that: In step S2, different phases of the MR image are registered, and digital subtraction processing is performed using the enhanced late arterial phase, portal venous phase, delayed phase and plain scan phase of the MR image to obtain a digital subtraction image.

8. A model for automatic detection and segmentation of active lesions after TACE surgery for liver cancer, characterized by: The construction method of claim 1 can input MR images of liver cancer patients after TACE surgery and output volume information of active lesions of liver cancer after TACE surgery.

Images