Pretreatment CT-based clinical deep learning radiomics model for predicting the response of hepatocellular carcinoma patients with portal vein thrombosis treated with TKIs plus PD-1 inhibitors and hepatic arterial infusion chemotherapy

Background This study aimed to construct a pretreatment contrast-enhanced CT (CECT)-based multimodal model for noninvasive forecasting of early treatment response in hepatocellular carcinoma (HCC) patients with portal vein tumor thrombosis (PVTT) who received hepatic arterial infusion chemotherapy combined with tyrosine kinase inhibitors and PD-1 inhibitors (HTI). Methods We retrospectively enrolled 116 patients with HCC and PVTT receiving HTIs (2019–2024), randomly divided 7:3 into training (n = 81) and test (n = 35) cohorts. Early objective response was assessed 6–8 weeks after treatment using modified RECIST. Radiomic features were extracted from three-dimensional volumes of interest of the primary tumor and PVTT; deep features from the largest axial tumor slice using ResNet101. Six models were developed; the final radiomic–deep learning–clinical (RCLD) model integrated out-of-fold component probabilities. Discrimination, calibration, clinical utility, and overall survival (OS) stratification were evaluated. Results Sixty-eight patients responded (complete response, n = 3; partial response, n = 65); the objective response rate was 58.6% and disease control rate 91.4%. Four radiomic and three deep learning features were selected; cirrhosis was the only clinical variable. RCLD achieved the highest AUC (training 0.965, 95% CI 0.929–1.000; test 0.924, 95% CI 0.822–1.000). In the test cohort, RCLD outperformed clinical-only and radiomics–clinical models ( p = 0.003 and p = 0.026), with no significant differences from other imaging-based models. RCLD stratified OS (training P = 0.003; test P = 0.027). Conclusion A pretreatment CECT-based multimodal model integrating radiomics, deep learning, and cirrhosis predicted early objective response and stratified OS risk in HCC with PVTT receiving HTIs. Prospective multicenter validation is required.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-19
DOI
https://doi.org/10.21203/rs.3.rs-11554224/v1
Primary Topic
Radiomics and Machine Learning in Medical Imaging
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Pretreatment CT-based clinical deep learning radiomics model for predicting the response of hepatocellular carcinoma patients with portal vein thrombosis treated with TKIs plus PD-1 inhibitors and hepatic arterial infusion chemotherapy

GuoRui Wang, Chunxue Wu, Jiadong Liu, Hao Zhang et al.
Research Square
Radiomics and Machine Learning in Medical Imaging
preprint

Pretreatment CT-based clinical deep learning radiomics model for predicting the response of hepatocellular carcinoma patients with portal vein thrombosis treated with TKIs plus PD-1 inhibitors and hepatic arterial infusion chemotherapy

GuoRui Wang, Chunxue Wu, Jiadong Liu, Hao Zhang, Jianjun Han, Bin Zhang, Hongtian Cheng, Fang Liu, Yimin Sun, Xiaochen Ren
preprint en

Abstract

Background This study aimed to construct a pretreatment contrast-enhanced CT (CECT)-based multimodal model for noninvasive forecasting of early treatment response in hepatocellular carcinoma (HCC) patients with portal vein tumor thrombosis (PVTT) who received hepatic arterial infusion chemotherapy combined with tyrosine kinase inhibitors and PD-1 inhibitors (HTI). Methods We retrospectively enrolled 116 patients with HCC and PVTT receiving HTIs (2019–2024), randomly divided 7:3 into training (n = 81) and test (n = 35) cohorts. Early objective response was assessed 6–8 weeks after treatment using modified RECIST. Radiomic features were extracted from three-dimensional volumes of interest of the primary tumor and PVTT; deep features from the largest axial tumor slice using ResNet101. Six models were developed; the final radiomic–deep learning–clinical (RCLD) model integrated out-of-fold component probabilities. Discrimination, calibration, clinical utility, and overall survival (OS) stratification were evaluated. Results Sixty-eight patients responded (complete response, n = 3; partial response, n = 65); the objective response rate was 58.6% and disease control rate 91.4%. Four radiomic and three deep learning features were selected; cirrhosis was the only clinical variable. RCLD achieved the highest AUC (training 0.965, 95% CI 0.929–1.000; test 0.924, 95% CI 0.822–1.000). In the test cohort, RCLD outperformed clinical-only and radiomics–clinical models ( p = 0.003 and p = 0.026), with no significant differences from other imaging-based models. RCLD stratified OS (training P = 0.003; test P = 0.027). Conclusion A pretreatment CECT-based multimodal model integrating radiomics, deep learning, and cirrhosis predicted early objective response and stratified OS risk in HCC with PVTT receiving HTIs. Prospective multicenter validation is required.

Research Square
Shandong Tumor Hospital (CN), Dongyang People's Hospital (CN), Shandong First Medical University (CN)
Radiomics and Machine Learning in Medical Imaging
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon