The Survival Analysis Of Hepatocellular Carcinoma Treated With ¹²⁵I Ra-Dioactive Seed Implantation Based On Radiomics

Objective:To develop a prediction model for survival after ¹²⁵I radioactive particle implantation in patients with hepatocellular carcinoma based on preoperative radiomics features combined with some clinical indicators.Methods:A retrospective analysis was performed on ¹²⁵I radioactive particle implantation in Nanchong Central Hospital for the treatment of liver cancer patients from January 2020 to January 2022 Preoperative clinical data and imaging data（34 cases in total）.It is based on two experienced radiologists through the Radcloud big data artificial intelligence research platform version2.1.2（developed by Beijing Huiying Medical Technology Company,website is http://radcloud.cn）.The patient’s preoperative portal venous stage CT image manually segmented the tumor sensing region（ROI）and extracted the radiomics Characteristics,one of the doctors repeated the work after a month.Within-group correlation coefficient（ICC）and Bland-Altman plot analysis were used to show inter-and intra-observer agreement and measurement variation.For radiomics,Variance Threshold（VT）and univariate feature selection were analyzed by Radcloud platform omics（SelectKBest）and the Minimum Absolute Shrinkage Algorithm（Lasso）implement radiomics feature dimensionality reduction.The univariate Cox proportional hazards regression model screened out the risk factors affecting the survival of patients with liver cancer treated with ¹²⁵I radioactive particle implantation.The above risk factors and patient survival outcomes were binary logistic regression to obtain a joint prediction model,and the univariate prediction model and multivariate prediction model were compared and analyzed by the subject’s working characteristic curve（ROC curve）,and the optimal threshold and corresponding specificity and sensitivity were found according to the Jordon index.A multivariate Cox risk proportional regression model was constructed,and a visualization of the survival model of multivariate Cox prediction ¹²⁵I radioactive particle implantation in liver cancer patients was completed in R（Nomogram）and demonstrate the clinical application of the model.Results:1.In this study,none of the 34 patients were lost to follow-up as of the time of follow-up.Sixteen of them died and the mean survival time for all patients was 13.265±6.473 months.2.The inter-observer and intra-observer ICC were 0.979（95%CI:0.959-0.990）and 0.998（95%CI:0.995-0.999）,respectively,and it could be observed via Bland-Altman plots that 32 scatter points were within the 95%CI interval,indicating good tumor ROI consistency for manual segmentation.3.A total of five statistically significant correlated factors were screened by the univariate Cox proportional hazards regression model:tumor maximum diameter,radiomics characteristics（gradient_firstorder_Energy（GFE）,original_shape_Maximum3DDiameter（OSM）,waveletLLH_firstorder Energy（WHFE）,waveletLLH_glszm_ZoneEntropy（WHGZ））risk ratio（HR）was 1.290（95%CI:1.129-1.473,p=0.000）,1.001（95%CI:1.000-1.002,p=0.014）,1.018（95%CI:1.005-1.031,p=0.007）,1.004（95%CI:1.001～1.006,p=0.004）、2.66（95%CI:1.038～6.816,p=0.042）;ROC curve analysis showed that the optimal univariate（tumor maximum diameter）prediction model threshold was 7.95 cm,AUC=0.898,sensitivity 0.750,specificity 0.944.The AUC=0.948,sensitivity 0.750,and specificity of the multivariate joint prediction model are more accurate than any one-factor model and are the best model.4.Multi-factor Cox proportional risk model analysis showed that tumor maximum diameter HR=1.308（95%CI:1.040～1.646,p=0.022）was an independent risk factor affecting the survival of patients with hepatocellular carcinoma treated with ¹²⁵I radioactive particle implantation.The above multifactorial Cox proportional risk prediction model was demonstrated to be of good clinical application by visualizing the nomogram.Conclusion:Patients with hepatocellular carcinoma with a maximum tumor diameter>7.95 cm had a lower survival rate after ¹²⁵I radioactive seed implantation.The combination of multiple indicators was more accurate in predicting patient survival than any single indicator.The Nomogram model based on a multifactorial Cox proportional risk regression of radiomics and some clinical indicators is important in predicting survival after ¹²⁵I radioactive seed implantation for hepatocellular carcinoma patients.