Search

CN-122023376-A - Early recurrence risk prediction method and system after liver transplantation of hepatocellular carcinoma based on multi-phase CT image

CN122023376ACN 122023376 ACN122023376 ACN 122023376ACN-122023376-A

Abstract

A method and a system for predicting early recurrence risk after liver transplantation of hepatocellular carcinoma based on multi-phase CT image belong to the technical field of medical image processing and artificial intelligence. The method comprises the steps of S1, multi-phase CT image acquisition and standardized pretreatment, S2, delineating a tumor area and a peritumor area to form a target volume, S3, extracting radiological features and deep learning features, S4, multi-round feature screening, constructing a deep learning radiological nomogram prediction model by fusing clinical risk factors, S5, risk prediction and layering, and S6, combining with a traditional clinical standard to realize incremental value integration. The system comprises an image access and preprocessing module, a feature calculation engine, a model management module, a risk assessment module and a clinical decision support module. By jointly analyzing the tumor and the peritumor area and fusing multidimensional features, the problems of insufficient information utilization, difficult clinical integration and the like in the prior art are solved, the early recurrence prediction precision is improved, and the support is provided for clinical decisions.

Inventors

  • WU ZIQIAN
  • REN KE

Assignees

  • 厦门大学

Dates

Publication Date
20260512
Application Date
20260212

Claims (10)

  1. 1. The method for predicting early recurrence risk after liver transplantation of hepatocellular carcinoma based on multi-phase CT image is characterized by comprising the following steps: S1, data acquisition and preprocessing, namely acquiring multi-phase enhanced CT images including a flat scanning phase, an arterial phase, a portal vein phase and a delay phase, which are acquired before liver transplantation of a patient with hepatocellular carcinoma, carrying out non-rigid registration on the images of the flat scanning phase, the arterial phase and the delay phase by taking the images of the portal vein phase as references, ensuring the spatial position consistency of each phase; S2, drawing and amplifying a target volume, namely manually drawing the whole tumor volume on the portal vein image to serve as a tumor area, and expanding the boundary of the tumor area by a preset width through morphological expansion operation to generate an annular tumor peripheral area, wherein the tumor area and the tumor peripheral area form the target volume; S3, extracting multi-dimensional image features: S31, extracting radiological features, namely respectively extracting first-order statistical features, three-dimensional shape features and texture features from the images of the horizontal scanning period, the arterial period, the portal vein period and the delay period of the target volume by using a feature extraction tool, wherein the texture features comprise features calculated from a gray level co-occurrence matrix, a gray level run-length matrix, a gray level size area matrix, an adjacent gray level difference matrix and a gray level dependency matrix, and the feature extraction is simultaneously carried out on an original image and a derivative image subjected to preprocessing by at least one of wavelet transformation, laplace Gaussian filtering, square root, logarithm, index, gradient, two-dimensional and three-dimensional local binary pattern filtering; S32, deep learning feature extraction, namely using a pre-trained deep residual convolutional neural network to forward propagate the multi-phase image of the target volume, and extracting an activation value before a network full-connection layer as the deep learning feature; S4, feature screening and model construction, namely firstly screening initial features which are obviously related to early recurrence by using Mann-Whitney U test on a training set, then removing highly redundant features by Spearman correlation analysis, finally carrying out final feature selection by adopting a minimum absolute shrinkage and selection operator logistic regression algorithm, and constructing a radiohistology score and a deep learning score based on the features; S5, risk prediction and application, namely inputting a multi-phase CT image of a patient to be detected into the prediction model to obtain early recurrence risk scores of the patient, and classifying the patient into a high risk group or a low risk group based on a preset risk score threshold.
  2. 2. The method according to claim 1, characterized in that in step S2, the preset width is 5mm.
  3. 3. The method according to claim 1, wherein in step S32, the depth residual convolutional neural network is ResNet-18 network, and the feature extraction comprises the specific steps of adjusting window width window level of CT image to soft tissue window, mapping gray value to [0, 255] interval, and inputting pre-trained ResNet-18 network.
  4. 4. The method of claim 1, wherein in step S4, the independent clinical risk factors include barcelona clinical liver cancer stage, microvascular invasion status, and periarterial aneurysm reinforcement.
  5. 5. The method of claim 1, wherein in step S4, the RAD-score and the DL-score are integrated by the deep-learning radiogram prediction model having a calculation formula of ：RAD-score = -1.75234564644120 + (0.346371678587235) × original_glszm_SmallAreaEmphasis.NP peritumor + (0.618500661255596) × original_glszm_SmallAreaHighGrayLevelEmphasis.AP peritumor + … + (0.646922842075291) × wavelet-HHH_firstorder_Kurtosis.VP tumor; and a calculation formula of ：DL-score = -0.241208216350357 + (0.32547062335402) × feature_1.AP peritumor + … + (0.0170101365332128)× feature_180.DP tumor;.
  6. 6. The method of claim 1, further comprising the step of step S6 of incremental value assessment and integration, combining the patient risk score obtained in step S5 with at least one conventional prognosis evaluation of hepatocellular carcinoma liver transplantation to form a composite risk stratification.
  7. 7. A post-liver transplantation early recurrence risk prediction system for hepatocellular carcinoma implementing the method of any one of claims 1-6, comprising: the image access and preprocessing module is used for executing image registration, resampling and target volume sketching and amplification; The feature calculation engine is integrated with a radiology feature extraction library and a pre-trained depth residual convolution neural network model and is used for executing the multi-dimensional image feature extraction; The model management module is internally provided with a feature selection algorithm and a logistic regression modeling unit and is used for storing and executing feature screening and model construction, and generating and maintaining the deep learning radiological alignment chart prediction model; And the risk assessment module is used for loading the prediction model, processing newly input CT image data, calculating and outputting individualized early recurrence risk scores and layering suggestions.
  8. 8. The system of claim 7, further comprising a clinical decision support module for automatically comparing and integrating the output of the risk assessment module with conventional prognostic assessment criteria results in a patient clinical database to generate a structured report comprising a composite risk stratification.
  9. 9. An electronic device comprising a processor and a memory, the memory storing a computer program, wherein the computer program when executed by the processor implements the method of any one of claims 1 to 6.
  10. 10. A computer readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, implements the method according to any of claims 1 to 6.

Description

Early recurrence risk prediction method and system after liver transplantation of hepatocellular carcinoma based on multi-phase CT image Technical Field The invention belongs to the technical field of medical image processing and artificial intelligence, and particularly relates to a method and a system for predicting early recurrence risk after liver transplantation of hepatocellular carcinoma based on multi-phase CT images by utilizing preoperative multi-phase computed tomography images, combining manually designed radiohistology features with data-driven deep learning features and constructing a nomogram model. Background Hepatocellular carcinoma is a common malignant tumor worldwide, and liver transplantation is one of the important radical therapeutic means. However, the scarcity of donor organs requires that the choice of grafting candidates must be optimized to ensure optimal post-operative outcome. At present, the Milan standard, the san Francisco calibration standard of the university of California and the like widely applied clinically are mainly based on macro morphological characteristics (such as number and size) of tumors, while the Metro-Ticket 2.0 model, the Hangzhou standard and the like try to integrate serum biomarkers such as alpha fetoprotein and the like. Nonetheless, about 10% -15% of patients develop tumor recurrence after liver transplantation, with early recurrence (defined as within 2 years of surgery) generally being more aggressive and patient prognosis being poor. Traditional prognostic evaluation systems fail to fully exploit the abundant quantitative information underlying medical images that reflects tumor intrinsic heterogeneity and tumor-host interface microenvironment interactions. Radiometric techniques, which enable high throughput extraction of large amounts of quantitative features from medical images, have proven potential in assessing pathological grading of HCC, microvascular invasion status, etc. Deep learning, particularly convolutional neural networks, can automatically learn and characterize complex, deep patterns in images. However, the prior art schemes have significant limitations, including firstly, most researches focused on the internal tumor region only, neglecting the peritumoral microenvironment with important biological significance, secondly, lacking an effective framework for systematically and complementarily fusing interpretable radiological features with deep learning features with high expression capability, and finally, the developed novel prediction model is usually evaluated as an independent tool, and fails to precisely quantify and realize the 'incremental value' of the novel prediction model relative to the existing mature clinical standard, so that clinical integration and application are difficult. Therefore, a high-precision prediction tool capable of integrating multi-dimensional image information and clinical information and seamlessly interfacing with the existing clinical decision process is needed. Therefore, a brand new technical scheme capable of integrating multidimensional image information and clinical information, rigorously repeatable in method, clear in clinical integration path and capable of remarkably improving early recurrence prediction precision is needed, the defect of the prior art is overcome, and accurate support is provided for prognosis evaluation and clinical decision of hepatocellular carcinoma liver transplantation patients. Disclosure of Invention The invention aims to provide a novel deep learning radiology alignment map system based on a multi-phase CT image and an application thereof, wherein the deep learning radiology alignment map system is based on a preoperative multi-phase CT image and aims to overcome the defects of insufficient information utilization, unreasonable feature fusion, difficult clinical integration and the like in the existing early recurrence risk prediction technology after liver transplantation of hepatocellular carcinoma. The method is characterized in that a multiscale and multisource information fusion deep learning radiological alignment chart prediction model is constructed, manual radiological characteristics and automatic extraction deep learning characteristics are fused through systematically analyzing tumors and peri-tumor areas, and a composite prediction tool which can be used independently and can effectively enhance traditional standards is finally formed by combining key clinical pathological factors, so that accurate prediction of early recurrence risk of a hepatocellular carcinoma patient after liver transplantation is realized. In order to achieve the above purpose, the invention adopts the following technical scheme: In a first aspect, a method for predicting early recurrence risk after liver transplantation of hepatocellular carcinoma based on multi-phase CT image is provided The method takes preoperative multi-phase enhanced CT images as a core data source, sys