Construction method and system of liver cancer radiofrequency ablation short-term curative effect and postoperative recurrence model

Abstract

The invention belongs to but is not limited to the technical field of medicine, and particularly relates to a method and system for constructing a liver cancer radiofrequency ablation short-term curative effect and postoperative recurrence model, a GE Resolution 256-row energy spectrum CT scanner is used for obtaining specific parameters of a focus area, and the correlation between multi-parameter data and liver cancer radiofrequency ablation postoperative short-term curative effect and postoperative recurrence is analyzed; the method comprises the following steps of: delineating a focus, extracting image feature texture information, analyzing the correlation between the image feature texture information and the liver cancer radiofrequency ablation postoperative recurrence rate, and discussing the application value of energy spectrum CT image radiomics in predicting the liver cancer radiofrequency ablation postoperative recurrence; clinical data including age, gender, laboratory examination, pathological types, size and number of lesions are collected and sorted, correlation analysis is carried out on the clinical data and relapse after radiofrequency ablation, and indexes related to relapse of liver cancer patients are screened; in combination with related clinical data and energy spectrum CT image radiomics quantitative characteristics, an optimal postoperative recurrence prediction model is established through an artificial intelligence method, and clinical verification is carried out.

Description

TECHNICAL FIELD

[0001] The application belongs to but is not limited to the technical field of medicine, and particularly relates to a method and system for constructing a short-term therapeutic effect and postoperative recurrence model of liver cancer radiofrequency ablation. BACKGROUND

[0002] Primary liver cancer is the third leading cause of cancer death worldwide. Each of the common clinical treatment methods has its own shortcomings. Radiofrequency ablation has the advantages of good curative effect, small trauma, simple operation, easy tolerance of patients, and fast recovery. However, radiofrequency ablation has a high tumor residual rate and recurrence rate after the operation. Therefore, how to effectively evaluate the short-term therapeutic effect and postoperative recurrence probability of liver cancer radiofrequency ablation is a key problem that needs to be solved in clinical practice. Studies have shown that spectral CT multi-parameter images can effectively detect micro lesions and are highly sensitive to hemorrhagic signals. The clinical characteristics of patients, the nature of tumors, and the effect of the operation are main factors leading to the postoperative recurrence of liver cancer radiofrequency ablation.

[0003] In a study published by Chunwang Yuan et al. in 2019, a recurrence prediction nomogram based on CT imaging features and clinical pathological factors was constructed to predict the early recurrence of hepatocellular carcinoma (HCC) patients after receiving radiofrequency ablation treatment. The study extracted 647 radiomic features from the three-phase images of enhanced CT, performed feature selection and model construction through the minimum-redundancy maximum-relevance method (mRMR) and LASSO Cox regression, and constructed a combination model that fused clinical and imaging biomarkers, which achieved a prediction performance of C-index about 0.755 in the validation cohort.

[0004] This method does not use spectral CT (Spectral CT) to obtain specific multi-parameters such as iodine concentration, nor does it use high-dimensional quantitative parameters such as spectral curve slope and standardized iodine concentration for model training, thus lacking the ability to specifically monitor the composition changes of lesions after radiofrequency ablation. In addition, the nomogram is mainly based on conventional CT imaging and clinical information, lacks the use of artificial intelligence deep models (such as Transformers) for multi-dimensional feature fusion and time series prediction, and limits the prediction accuracy and dynamic response capability, which cannot provide quantitative judgment of short-term therapeutic effect and adaptive iteration ability of the model after the operation. SUMMARY

[0005] In view of the problems in the prior art, the application provides a method and system for constructing a short-term therapeutic effect and postoperative recurrence model of liver cancer radiofrequency ablation.

[0006] The application is implemented as follows. A method for constructing a short-term therapeutic effect and postoperative recurrence model of liver cancer radiofrequency ablation comprises the following steps.

[0007] S1, Acquisition of spectral multi-parameter data: Using a GE Revolution 256-slice spectral CT scanner, specific parameters of the lesion area are acquired, including lesion size and shape, lesion iodine uptake ratio, lesion standardized iodine concentration, and spectral curve slope. The correlation between multi-parameter data and short-term efficacy after radiofrequency ablation for liver cancer and postoperative recurrence is analyzed.

[0008] S2, Quantitative feature extraction of spectral CT image radiomics: The lesion is outlined, and image texture information is extracted. The correlation between the recurrence rate after radiofrequency ablation for liver cancer and the image texture information is analyzed, and the application value of spectral CT image radiomics in predicting the recurrence after radiofrequency ablation for liver cancer is explored.

[0009] S3, Correlation between multi-dimensional data and recurrence after radiofrequency ablation for liver cancer: Clinical data, including age, gender, laboratory examination, pathological type, lesion size and number, are collected and analyzed for correlation with recurrence after radiofrequency ablation. The recurrence-related indicators for liver cancer patients are screened.

[0010] S4, Establishment of a model for predicting recurrence after radiofrequency ablation for liver cancer: Based on relevant clinical data and quantitative features of spectral CT image radiomics, the best postoperative recurrence prediction model is established by artificial intelligence method, and clinical verification is performed.

[0011] Further, the method for constructing a short-term efficacy and postoperative recurrence model for radiofrequency ablation of liver cancer further comprises:

[0012] Evaluating the short-term efficacy of radiofrequency ablation: Using spectral CT technology, the material composition of the lesion after radiofrequency ablation is quantitatively analyzed by multiple parameters to accurately determine whether the ablation is complete, thereby improving the ablation rate and providing an image-based quantitative basis for guiding precise ablation in clinical practice.

[0013] Constructing a prediction model for recurrence after radiofrequency ablation: Collect and analyze radiofrequency ablation-related data for liver cancer, screen risk factors and characteristic parameters for liver cancer recurrence, and construct a prediction model.

[0014] Further, the research subjects include: liver cancer patients who underwent radiofrequency ablation.

[0015] Inclusion criteria: (1) The diameter of liver cancer tumor is not more than 5 cm, and the number is not more than 3; (2) Child-Pugh score is A or B; (3) Complete clinical data; (4) First treatment, and the treatment method is spectral CT-guided radiofrequency ablation puncture, and spectral enhancement CT scan is performed immediately after the operation.

[0016] Exclusion criteria: (1) History of previous treatment; (2) Liver cancer patients with metastasis or combined with other tumors; (3) Patients who cannot tolerate treatment due to other reasons.

[0017] Further, the observation indicators specifically include:

[0018] (1)Preoperative history of hepatitis B, alpha-fetoprotein, Child-Pugh classification, clinical liver cancer stage, platelet count;

[0019] (2)Imaging indicators, including: tumor number, tumor long diameter, tumor location, cirrhosis.

[0020] Further, the spectral CT scan: take a supine position, raise both hands, the scanning range is from 3cm above the diaphragm to the lower pole of the kidney, and a double-barreled high-pressure injector is connected. The spectral plain scan parameters adopt the GSI sequence: kV: 80-140; mA: 315; detector width: 80mm; the reconstruction layer thickness and interval are both 0.625mm; standard reconstruction; the spectral reconstruction parameters: GSI image type: kVp-Like; Data File: on; ASiR is set to 50%; the enhancement scan adopts the intelligent tracking method, the threshold is set to 150Hu, the monitoring starts 10s after the injection of contrast agent, and when the threshold exceeds the set range, the scanning is started 8s later. The arterial phase scanning is completed, and the portal vein phase scanning is performed 25s later. The portal vein phase scanning is completed, and the delay phase scanning is performed 60s later. The scanning parameters and the reconstruction parameters are the same as those of the plain scan. The radiation dose DLP of the whole scan is not more than 1800mGy.cm.

[0021] Further, the spectral data post-processing includes: lesion selection and labeling: after the scanning is completed, it is transmitted to the GE AW4.7 post-processing workstation, the spectral data package is selected for post-processing, and the regions of interest are placed on the iodine water images in the arterial phase and the portal vein phase respectively. When the lesion is outlined, the complete lesion needs to be included, while avoiding the surrounding blood vessels. The ROI of the liver parenchyma needs to avoid the lesion, the bile duct and the blood vessels, and the ROI of the aorta is smaller than the blood vessel lumen area. The iodine concentration and the water concentration in the arterial phase and the portal vein phase are measured, and the lesion iodine uptake ratio IUR=IC lesion / IC is calculated. The normalized iodine concentration of the liver lesion NIC=IC lesion / IC aorta.

[0022] Another object of the present application is to provide a liver cancer radiofrequency ablation short-term efficacy and postoperative recurrence model construction system for realizing the liver cancer radiofrequency ablation short-term efficacy and postoperative recurrence model construction method, comprising:

[0023] The spectral multi-parameter data acquisition module: using a GE Revolution 256-row spectral CT scanner, specific parameters of the lesion area are acquired, including lesion size and shape, lesion iodine uptake ratio, lesion normalized iodine concentration, spectral curve slope, and the correlation of the multi-parameter data with the short-term efficacy and postoperative recurrence after liver cancer radiofrequency ablation is analyzed.

[0024] Spectrum CT image imageomics quantitative feature extraction research module: draw the lesion, extract the image feature texture information, analyze the correlation thereof with the recurrence rate after radiofrequency ablation of liver cancer, and explore the application value of spectrum CT image imageomics in predicting the recurrence after radiofrequency ablation of liver cancer;

[0025] Multi-dimensional data and recurrence after radiofrequency ablation of liver cancer correlation research module: collect and arrange clinical data, including age, gender, laboratory examination, pathological type, lesion size and quantity, and perform correlation analysis with the recurrence after radiofrequency ablation, and screen the recurrence-related indexes of liver cancer patients;

[0026] Prediction model of recurrence after radiofrequency ablation of liver cancer establishment module: in combination with relevant clinical data and spectrum CT image imageomics quantitative features, the best postoperative recurrence prediction model is established through an artificial intelligence method, and clinical verification is performed.

[0027] Another object of the present application is to provide a computer device comprising a memory and a processor, the memory storing a computer program, the computer program being executed by the processor to make the processor execute the steps of the method for constructing the short-term efficacy and postoperative recurrence model of radiofrequency ablation of liver cancer.

[0028] Another object of the present application is to provide a computer-readable storage medium storing a computer program, the computer program being executed by the processor to make the processor execute the steps of the method for constructing the short-term efficacy and postoperative recurrence model of radiofrequency ablation of liver cancer.

[0029] Another object of the present application is to provide an information data processing terminal comprising the construction system of the short-term efficacy and postoperative recurrence model of radiofrequency ablation of liver cancer.

[0030] In combination with the above technical solutions and solved technical problems, the technical solution to be protected by the present application has the following advantages and positive effects:

[0031] The present application evaluates the short-term efficacy after radiofrequency ablation by spectrum CT multi-parameter imaging, mines deep texture features of images based on imageomics method, comprehensively analyzes multi-dimensional data, screens characteristic parameters that can represent the recurrence after radiofrequency ablation of liver cancer, and then establishes a prediction model, thereby providing a new quantitative analysis method for immediate effect evaluation and postoperative recurrence prediction of radiofrequency ablation of liver cancer, and providing an important reference basis for clinical diagnosis and treatment decision-making. Most of the current researches are follow-up of postoperative recurrence, and the immediate effect evaluation of radiofrequency ablation of liver cancer is often ignored. The present research combines the immediate effect evaluation and postoperative recurrence features of radiofrequency ablation of liver cancer, and can more comprehensively establish a prediction model. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1A flow chart of a method for constructing a short-term efficacy and postoperative recurrence model of liver cancer radiofrequency ablation provided by the embodiment of the present application;

[0033] Figure 2 A technical roadmap for preventing the method for constructing a short-term efficacy and postoperative recurrence model of liver cancer radiofrequency ablation provided by the embodiment of the present application;

[0034] Figure 3 An arterial phase spectral image and a ROI placement position map provided by the embodiment of the present application;

[0035] Figure 4 A venous phase spectral image provided by the embodiment of the present application;

[0036] Figure 5 A positioning map in liver cancer radiofrequency ablation provided by the embodiment of the present application;

[0037] Figure 6 A postoperative spectral CT image of liver cancer radiofrequency ablation provided by the embodiment of the present application;

[0038] Figure 7 A postoperative spectral processing image of liver cancer radiofrequency ablation provided by the embodiment of the present application;

[0039] Figure 8 A postoperative review spectral CT image of liver cancer radiofrequency ablation provided by the embodiment of the present application;

[0040] Figure 9 A system structure diagram of the method for constructing a short-term efficacy and postoperative recurrence model of liver cancer radiofrequency ablation provided by the embodiment of the present application. DETAILED DESCRIPTION

[0041] In order to make the object, technical scheme and advantages of the present application clearer, the present application will be further described in detail below with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0042] The present application uses spectral CT technology, combines imageomics and clinical multidimensional data, and explores a prediction model for short-term efficacy and recurrence after liver cancer radiofrequency ablation. Through advanced imaging technology and artificial intelligence modeling means, it is expected to provide scientific basis for accurate judgment and early intervention for the clinic, so as to improve the long-term benefit of patients.

[0043] First, the lesion spectral parameters are collected by GE Revolution 256-row spectral CT, including lesion size, shape, iodine uptake ratio, standardized iodine concentration and spectral curve slope, etc., and the lesion texture features are extracted by combining the imageomics platform to quantitatively evaluate the changes in the ablation area. This process not only can determine whether the ablation is complete, but also can provide key indicators for subsequent recurrence risk stratification.

[0044] Further integration of patient clinical data, such as age, gender, laboratory test results, liver function, tumor markers, pathological type, and tumor number and diameter, was conducted using statistical and machine learning methods to analyze their correlation with recurrence. Joint analysis of multi-source data helps screen for high-risk factors and construct a key feature set for predicting recurrence risk.

[0045] The study included 100 liver cancer patients who underwent radiofrequency ablation at our hospital. All patients met the inclusion criteria of tumor diameter not exceeding 5 cm, number not exceeding 3 tumors, and Child-Pugh score A or B, and were excluded if they had a history of previous treatment, metastasis, or other concurrent tumors. Immediately after the procedure, enhanced spectral CT scans were performed to ensure the completeness and consistency of ablation efficacy assessment and data acquisition.

[0046] By integrating and analyzing the above data, a predictive model based on artificial intelligence methods will be established to accurately predict recurrence after radiofrequency ablation for liver cancer. After clinical validation, this model can provide doctors with quantitative data for postoperative follow-up and treatment decisions, enabling individualized management and precision medicine, and promoting the transformation of minimally invasive liver cancer treatment from experience-based to data-driven approaches.

[0047] In current clinical practice, radiofrequency ablation for liver cancer, as a minimally invasive local treatment, has become an important option for patients unsuitable for surgical resection. However, postoperative efficacy assessment and recurrence prediction remain core technical challenges in its application. Conventional imaging methods largely rely on two-dimensional CT or MRI for intuitive judgment of lesion size and morphology. This approach is heavily influenced by the operator's experience and subjective interpretation of images, making it difficult to objectively quantify ablation integrity and potential residual lesions. This results in some patients experiencing recurrence during early follow-up examinations, limiting the promotion and efficacy of radiofrequency ablation.

[0048] To address this issue, spectral CT technology offers new possibilities. Compared to traditional imaging, spectral CT can acquire information on tissue decomposition, such as iodine concentration, the slope of the spectral curve, and the iodine uptake ratio. These parameters reflect the blood supply characteristics and tissue composition differences of the lesion. By comparing the spectral parameters of the ablation zone with the surrounding liver parenchyma, it is possible to more accurately identify whether the ablation range is complete, especially in the context of cirrhosis, avoiding errors caused by relying solely on morphological judgment. This multi-parameter quantitative method provides an objective and repeatable means of efficacy evaluation for industrial applications.

[0049] Meanwhile, the development of radiomics has expanded the prediction of liver cancer recurrence beyond visually perceptible indicators. By extracting and modeling hundreds of features from spectral CT images, such as texture, gray-level co-occurrence matrix, and shape complexity, it is possible to capture the heterogeneity within lesions. This information is often closely related to tumor biological behavior. Radiomics can reveal early signals that are difficult to detect under traditional imaging, thereby enabling multi-level prediction from macro to micro in industrial applications and improving the scientific rigor of recurrence risk stratification.

[0050] In terms of data, this study not only relies on imaging but also integrates clinical data, including patients' age, gender, history of viral hepatitis, tumor marker levels, liver function indicators, and pathological types. This information, together with radiomics features, constitutes a multimodal database. Through statistical screening and training with artificial intelligence algorithms, key variables for recurrence risk can be identified. The industrial value lies in the fact that this method can establish a unified indicator system, reduce reliance on manual intervention, and achieve standardized predictions that can be transferred between hospitals.

[0051] The key to its working principle lies in the deep fusion of multi-source data. Spectral CT parameters reflect material composition, radiomics reveals tissue heterogeneity, and clinical indicators reflect the body's state. These three elements are processed through feature selection, dimensionality reduction, and nonlinear fitting by an artificial intelligence model to ultimately form a multi-parameter model for predicting recurrence. This model can generate a recurrence risk score immediately after surgery, helping doctors develop individualized follow-up and retreatment strategies, thus overcoming the shortcomings of existing technologies that rely on single indicators and experience-based judgment.

[0052] The method proposed in this study addresses two core issues in industrial applications: first, it enables objective quantitative assessment of the short-term efficacy of radiofrequency ablation for liver cancer, improving the accuracy of immediate post-operative assessment; second, it establishes a multimodal model for recurrence prediction, allowing clinicians to perform risk stratification immediately after treatment. This technological approach not only has clear medical value but also provides practical technical support for the development of artificial intelligence in medical imaging and precision medicine.

[0053] Child-Pugh Classification Standards

[0054]

[0055]

[0056] Note: Child-Pugh Level A: 5-6 points; Level B: 7-9 points; Level C: ≥10 points

[0057] Spectral CT Scan: The study was authorized by the hospital's ethics committee, and participants signed informed consent forms and contrast-enhanced CT consent forms. Participants were placed in a supine position with their arms raised. The scanning range extended from 3 cm above the diaphragm to the lower pole of the kidney, using a dual-barrel high-pressure injector. The plain spectral scan parameters used the GSI sequence: kV: 80-140; mA: 315; detector width: 80 mm; reconstruction slice thickness and interslice spacing were both 0.625 mm; standard reconstruction. Spectral reconstruction parameters: GSI image type: kVp-Like; Data File: on; ASiR set to 50%. The enhanced scan used a smart tracking method, monitoring the portal vein level. The threshold was set to 150 HU. Monitoring began 10 seconds after contrast agent injection, and scanning started 8 seconds after the threshold exceeded the set range. The portal vein phase scan was performed 25 seconds after the arterial phase scan, and the delayed phase scan was performed 60 seconds after the portal vein phase scan. The scanning and reconstruction parameters were the same as the plain scan. The total radiation dose (DLP) for the entire scan did not exceed 1800 mGy·cm (approximately the 50th percentile nationally).

[0058] Post-processing of spectral data: Lesion selection and annotation: After scanning, the data is transferred to the GE AW4.7 post-processing workstation. The spectral data package is selected for post-processing. Regions of interest (ROIs) are placed on the iodine water maps of the arterial and portal venous phases, respectively. When delineating lesions, the entire lesion (including necrosis, hemorrhage, and edema within the tumor) must be included, while avoiding surrounding blood vessels. For liver parenchyma, ROIs must avoid lesions, bile ducts, and blood vessels. For the aorta, the ROI should be smaller than the lumen area of ​​the blood vessel (e.g., ...). Figure 3 , Figure 4 (As shown). Iodine (water) concentration (IC) and water (iodine) concentration (WC) were measured during the arterial and portal venous phases. The iodine uptake ratio of lesions (IUR) was calculated: IUR = IC lesion / IC; the normalizing iodine concentration of liver lesions (NIC) was calculated: NIC = IC lesion / IC aorta.

[0059] Application of Spectral CT in Radiofrequency Ablation of Liver Cancer: Spectral CT images (metal artifact removal techniques, MARs) provide a clearer view of the needle tip position compared to conventional CT, aiding in puncture localization. Figure 5 ).

[0060] A study on the immediate effect evaluation of radiofrequency ablation for liver cancer using spectral CT can effectively identify the liquefaction area and the surrounding ring-shaped enhanced edema zone after ablation by separating iodine and water images. Figure 6 Post-processing of images using energy dispersive spectroscopy can effectively identify the correlation between lesions and normal liver parenchyma, thereby accurately assessing whether ablation is complete.Figure 7 ).

[0061] Figure 6 Image a is the water image from the radiofrequency ablation procedure for liver cancer, image b is the iodine image from the radiofrequency ablation procedure, and image cd is the pseudo-color image. The lesion is clearly displayed.

[0062] Figure 7 After radiofrequency ablation of liver cancer, spectral images were processed with regions of interest (ROIs) placed in the lesion area and the liver parenchyma area. Image a is the arterial phase enhanced image, image b is the effective atomic number map of the spectral spectrum, and image c is the slope map of the spectral curve. This image shows that the two have similar origins and the ablation was complete.

[0063] A study on the use of spectral CT in follow-up after radiofrequency ablation for liver cancer: spectral CT monographs can optimize image quality and improve the signal-to-noise ratio. Retrospective analysis of spectral CT images after radiofrequency ablation for liver cancer revealed that monographs with energy levels of 60-80 keV produced the best image quality. Figure 8 ). Figure 8 Postoperative radiofrequency ablation of liver cancer followed by follow-up spectral CT images: Image a is a mixed-energy image from conventional CT, and Image b is a single-energy image (60 keV) from spectral CT. Comparison of the two images shows that spectral CT images have higher resolution, higher contrast of lesions, clearer display, and are more effective in determining whether there is recurrence compared to conventional CT.

[0064] like Figure 9 As shown, the system for constructing a model of short-term efficacy and postoperative recurrence of radiofrequency ablation for liver cancer provided in this embodiment of the invention includes:

[0065] Spectral multi-parameter data acquisition module: Using a GE Revolution 256-slice spectral CT scanner, specific parameters of the lesion region are acquired, including lesion size and shape, lesion iodine uptake ratio, lesion standardized iodine concentration, and spectral curve slope. The correlation between multi-parameter data and short-term efficacy and postoperative recurrence after radiofrequency ablation of liver cancer is analyzed.

[0066] The quantitative feature extraction module of spectral CT image radiomics: delineate lesions, extract image feature texture information, analyze its correlation with the recurrence rate after radiofrequency ablation of liver cancer, and explore the application value of spectral CT image radiomics in predicting the recurrence of liver cancer after radiofrequency ablation.

[0067] The module for studying the correlation between multidimensional data and recurrence after radiofrequency ablation of liver cancer includes: collecting and organizing clinical data, including age, gender, laboratory tests, pathological type, lesion size and number, and conducting correlation analysis with recurrence after radiofrequency ablation to screen indicators related to recurrence in liver cancer patients.

[0068] The module for establishing a model to predict recurrence after radiofrequency ablation of liver cancer: Combining relevant clinical data and quantitative features of radiomics in spectral CT images, an optimal model for predicting postoperative recurrence is established using artificial intelligence methods and then clinically validated.

[0069] Example 1 (Overall Process Method)

[0070] In a top-tier hospital, 50 patients with liver cancer undergoing radiofrequency ablation were included. A GE Revolution 256-slice spectral CT scanner was used for pre- and post-operative scans. Acquired lesion parameters included size, morphology, iodine uptake ratio, standardized iodine concentration, and the slope of the spectral curve, which were then transmitted to a back-end storage platform.

[0071] By combining patients' clinical data such as age, gender, laboratory tests, and pathological type, an artificial intelligence algorithm was used to build a predictive model, which outputs the recurrence probability. The results showed that the model integrating clinical and radiomics data achieved an AUC of 0.89 in predicting postoperative recurrence, which was significantly higher than that of single-parameter prediction.

[0072] Example 2 (Short-term efficacy evaluation)

[0073] In the immediate images following radiofrequency ablation in 30 patients with hepatocellular carcinoma, iodine and water concentrations in the lesions were measured during the arterial and portal venous phases. The presence of residual enhancement areas in the lesions was determined by calculating the iodine uptake ratio and standardized iodine concentration.

[0074] The results showed that the iodine uptake ratio of the lesions in the complete ablation group was less than 0.2, while that in the incomplete ablation group was greater than 0.3. This significant difference can provide quantitative imaging evidence for evaluating short-term efficacy after surgery.

[0075] Example 3 (Radiomics Feature Extraction)

[0076] In arterial phase images of spectral CT, lesion ROIs are delineated, and texture features, including gray-level co-occurrence matrix energy, entropy, contrast, and homogeneity, are extracted. Morphological parameters such as volume and aspect ratio are also extracted.

[0077] The analysis results showed that the lesion texture entropy value of patients in the high recurrence group was significantly higher than that in the low recurrence group, and the volume irregularity index was also positively correlated with the risk of recurrence, indicating that radiomics features have independent value in predicting recurrence.

[0078] Example 4 (Feature Normalization and Screening)

[0079] Radiomic features of 80 patients were normalized, with all parameters scaled to between 0 and 1. A feature importance ranking algorithm was used to select the top 30 features highly correlated with recurrence.

[0080] The results show that texture entropy, gray-scale difference index, and energy spectrum curve slope are important features, and the prediction accuracy of the screened model is improved by 12% on the independent validation set.

[0081] Example 5 (Multidimensional Data Fusion Prediction)

[0082] In a prospective study, 100 patients were included, and their clinical data (history of hepatitis B, AFP level, platelet count, etc.) were extracted and constructed into structured feature vectors; at the same time, energy spectrum radiomics features were extracted to form high-dimensional texture vectors.

[0083] The two types of vectors were combined and input into the artificial intelligence model through a feature fusion layer. The results showed that the fusion model had an accuracy of 83% in predicting recurrence 1 year after surgery, which was better than the clinical model alone (71%) and the imaging model alone (74%).

[0084] Example 6 (AI Modeling Details)

[0085] A Transformer network is used as the prediction model, with the fused feature vector as input and the recurrence probability as output. Cross-entropy loss is used during training, and class weights are introduced when there is class imbalance.

[0086] On the validation set, the recurrence probability threshold was set to 0.5; values ​​above this were considered high-risk. Results showed that the model had a sensitivity of 85% and a specificity of 82%, effectively distinguishing between high-risk and low-risk patients.

[0087] Example 7 (Energy Spectrum Data Processing)

[0088] After scanning, the data was transferred to a GE AW4.7 post-processing workstation. The lesion ROI and aortic ROI were placed on the arterial and portal venous phase iodine maps, respectively, and the iodine concentration in the lesion and aorta was measured.

[0089] The calculated iodine uptake ratio at the lesion site was the iodine concentration at the lesion site divided by the iodine concentration in the liver parenchyma, and the standardized iodine concentration was the iodine concentration at the lesion site divided by the iodine concentration in the aorta. Both of these indicators were significantly correlated with the recurrence rate.

[0090] Example 8 (ROI Drawing Specifications)

[0091] Two experienced radiologists independently drew the ROIs. The ROI for the lesion covered the complete tumor boundary and avoided the surrounding blood vessels. The ROI for the liver parenchyma was selected in the area away from the lesion and bile duct.

[0092] Comparative analysis showed that standardized ROI plotting reduced the differences in repeated measurements, and the inter-group coefficient of variation for iodine concentration parameters decreased from 12% to 4%, significantly improving data stability.

[0093] Example 9 (System Implementation)

[0094] A system was constructed that includes a data acquisition module, a radiomics extraction module, a clinical data collection module, and a predictive model module. The data acquisition module automatically connects to the hospital's PACS and LIS systems to achieve automated data flow.

[0095] The system can complete predictive analysis within 3 hours after a patient completes radiofrequency ablation and push the recurrence risk grading results to the clinical terminal for doctors' reference, thereby improving the efficiency of clinical decision-making.

[0096] Example 10 (Model Validation and Iteration)

[0097] In a retrospective study, 200 patients were selected for model training, and another 50 independent cases were selected for validation. The accuracy rate of the predictive model results compared to actual follow-up recurrence outcomes was 81%.

[0098] The system records misclassified cases and feeds them back to the iteration unit, which automatically adjusts the model weights based on the feedback. After the update, the accuracy of the new validation set improved to 86%, demonstrating the model's ability to continuously optimize.

[0099] Evidence related to the technical effects obtained by the embodiments of the present invention.

[0100] Previous team research: Factor analysis of hepatocellular carcinoma recurrence after TACE: Forty-five patients with hepatocellular carcinoma who underwent TACE and were followed up for 6 months were collected. Among them, 26 cases recurred (recurrence group) and 19 cases did not recur (non-recurrence group). The tumor marker alpha-fetoprotein (AFP) level ≥400 ng / ml was significantly higher in the recurrence group than in the non-recurrence group (P < 0.05). Alanine aminotransferase (ALT) levels also showed a statistically significant difference (P < 0.05). The non-recurrence group had a higher degree of tumor differentiation. There were no statistically significant differences between the two groups in age, sex, comorbidities, Child-Pugh liver function classification, BCLC stage, AST, total bilirubin, and prothrombin time (P > 0.05) (Table 1). Multivariate logistic regression analysis showed that the proportion of patients with AFP > 400 ng / ml, tumor differentiation degree, number of lesions, iodized oil deposition effect, and whether the lesions showed enhancement were independent factors for early recurrence of hepatocellular carcinoma after TACE (P > 0.05) (Table 2).

[0101] Table 1

[0102]

[0103]

[0104] Table 2

[0105]

[0106] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0107] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for constructing a model of short-term therapeutic effect and postoperative recurrence of radiofrequency ablation of liver cancer, characterized in that, Comprise: S1, using spectral CT scanning equipment to obtain multi-parameter data of the lesion area, the parameters including lesion size, lesion morphology, lesion iodine uptake ratio, lesion standardized iodine concentration and spectral curve slope; S2, imageomics processing of spectral CT image, extracting lesion texture features, and analyzing its correlation with postoperative recurrence rate; S3, combined with the patient's clinical data, including age, gender, laboratory examination results, pathological type, lesion size and quantity, an index system related to postoperative recurrence is established; S4, based on artificial intelligence method, the clinical data and imageomics features are fused to establish a postoperative recurrence prediction model, and verified by clinical data.

2. The method of claim 1, wherein, Also includes: using spectral CT multi-parameter quantitative analysis of the material composition of the lesion after radiofrequency ablation, judging whether the ablation is complete by the change of iodine concentration, water concentration and lesion iodine uptake ratio, for evaluating short-term efficacy. 3.A method for extracting features of liver cancer recurrence after surgery based on spectral CT imaging, characterized in that, Comprise: In the arterial phase and portal phase iodine water image, the lesion area is outlined, the lesion texture, gray level co-occurrence matrix feature, morphological feature and differential based change trend feature are extracted, and the imageomics feature set is constructed.

4. The method of claim 3, wherein, The extracted imageomics features are normalized by min-max normalization, and the feature importance sorting method is used to select a feature subset highly related to postoperative recurrence.

5. A multi-dimensional data fusion prediction method for recurrence after liver cancer radiofrequency ablation, characterized in that, Comprise: The patient's clinical data is converted into a structured feature vector; The imageomics features are converted into a high-dimensional texture vector; The structured feature vector and the high-dimensional texture vector are merged through the feature fusion layer and input into the artificial intelligence model for prediction output.

6. The method of claim 5, wherein, The artificial intelligence model adopts the Transformer neural network structure, the input end is the fusion feature vector, and the output end is the recurrence probability. When the recurrence probability is greater than the threshold, it is determined as a high-risk patient.

7. A method for post-processing of liver cancer spectral CT data, characterized in that, Comprise: Place the lesion region of interest and the aorta region of interest on the arterial phase and portal phase iodine water images respectively, and measure the lesion iodine concentration and the aorta iodine concentration; Calculate the lesion iodine uptake ratio, which is the ratio of lesion iodine concentration to liver parenchyma iodine concentration; Calculate the lesion standardized iodine concentration, which is the ratio of lesion iodine concentration to aorta iodine concentration.

8. The method of claim 7, wherein, The lesion region of interest needs to cover the complete lesion, avoiding blood vessels and bile ducts, and the liver parenchyma region of interest needs to avoid the lesion and blood vessels.

9. A system for realizing short-term efficacy and postoperative recurrence prediction of radiofrequency ablation of liver cancer, characterized in that, Comprise: Spectral multi-parameter data acquisition module for collecting lesion size, shape, iodine uptake ratio, standardized iodine concentration and spectral curve slope; Imageomics feature extraction module for outlining the lesion and extracting texture features and differential features; Clinical data acquisition module for collecting age, gender, laboratory examination, pathological type, lesion size and quantity; Prediction model establishment module for fusing imageomics features and clinical features and generating recurrence risk assessment results through artificial intelligence algorithm.

10. The system of claim 9, wherein, The prediction model establishment module further comprises a verification unit and an iteration unit, the verification unit is used to compare the prediction results with the true recurrence results, and the iteration unit is used to update the model parameters based on the verification feedback.

Images