Background: Transcatheter arterial chemoembolization (TACE) is a mainstay treatment for intermediate and advanced hepatocellular carcinoma (HCC), with the potential to enhance patient survival. Preoperative prediction of postoperative response to TACE in patients with HCC is crucial.
Purpose: To develop a deep neural network (DNN)-based nomogram for the non-invasive and precise prediction of TACE response in patients with HCC.
Material and methods: We retrospectively collected clinical and imaging data from 110 patients with HCC who underwent TACE surgery. Radiomics features were extracted from specific imaging methods. We employed conventional machine-learning algorithms and a DNN-based model to construct predictive probabilities (RScore). Logistic regression helped identify independent clinical risk factors, which were integrated with RScore to create a nomogram. We evaluated diagnostic performance using various metrics.
Results: Among the radiomics models, the DNN_LASSO-based one demonstrated the highest predictive accuracy (area under the curve [AUC] = 0.847, sensitivity = 0.892, specificity = 0.791). Peritumoral enhancement and alkaline phosphatase were identified as independent risk factors. Combining RScore with these clinical factors, a DNN-based nomogram exhibited superior predictive performance (AUC = 0.871, sensitivity = 0.844, specificity = 0.873).
Conclusion: In this study, we successfully developed a deep learning-based nomogram that can noninvasively and accurately predict TACE response in patients with HCC, offering significant potential for improving the clinical management of HCC.
Keywords: Transcatheter arterial chemoembolization; deep neural network; hepatocellular carcinoma; radiomics; response.