关键词: 肝硬化, 肝储备功能, 影像组学, 体层摄影术, X线计算机

Abstract: Objective To investigate the feasibility of developing an image omics model based on portal phase CT images to extract features from the whole liver and specific liver segments. The Objective is to assess liver reserve function in cirrhotic patients with varying Child-Turcotte-Pugh (CTP) scores. Additionally, the study compares the diagnostic efficacy of the two models. Methods A retrospective analysis was conducted on 148 patients diagnosed with cirrhosis, categorized into CTP A and CTP B+C groups based on CTP scores. All patients were randomly divided into a training set (n=118) and a validation set (n=30). Regions of interest(ROI_t and ROI_s)were defined according to the segmentation of the whole liver and individual liver segements, respectively. Features were extracted and screened using the Spearman correlation coefficient and the least absolute shrinkage and selection operator (LASSO) for dimensionality reduction. The efficiency of the ROI_t and ROI_s models was evaluated using reveiver operating characteristic(ROC) curves and decision curve analysis(DCA). The Delong test was employed to compare the ROC curve efficiencies of the two models. Results The ROI_t model was established by selecting the 10 most significant features. For the training set, the area under the curve(AUC ) was 0.886(95%CI: 0.821 ~ 0.950), with an accuracy of 0.828, sensitivity of 0.849, and specificity of 0.826. For the test, the AUC was 0.895(95%CI: 0.797 ~ 0.992), with an accuracy of, 0.816, sensitivity of 0.884, and specificity of 0.870. The ROI_s model was established by selecting the 13 most significant features. For the training set, the AUC was 0.854(95%CI: 0.765 ~ 0.943), with an accuracy of 0.743, sensitivity of 0.800, and specificity of 0.853. For the test set,the AUC was 0.700(95%CI: 0.575-0.823), with an accuracy of 0.608, sensitivity of 0.744, and specificity of 0.706, respectively. The DeLong test indicated a statistically significant difference bewteen the ROC curves of the ROI_t and ROI_s models(P=0.0369). The DCA demonstrated that both the ROI_t and ROI_s models had good predictive ability for CTP classification, with the ROI_t model outperforming .the ROI_s model. Conclusion Based on the most significant features extracted from the whole liver and local regions of interest in the portal phase CT images of cirrhotic patients, both imaging omics models established using the LightGBM classifier can effectively evaluate liver reserve function in patients with different CTP grades. Notably, the imaging omics model based on regions of interest in the whole liver demonstrates superior performance.

Key words: Cirrhosis, Liver reserve function, Imaging omics, Tomography, X-ray computer