A machine learning algorithm facilitates treatment selection for up to three hepatocellular carcinomas ≤ 3 cm in size

Liver resection (LR) and radiofrequency ablation (RFA) are recommended for patients with up to three hepatocellular carcinomas (HCCs) measuring ≤ 3 cm and preserved liver function. Nevertheless, treatment allocation is decided according to the preference of the treating doctor. This study aimed to develop a predictive model to guide treatment decisions based on survival outcomes. A total of 18,958 patients with HCC with up to three nodules measuring ≤ 3 cm were included from the nationwide survey of Japan. Deep survival analysis was performed using the Recurrent Deep Survival Machines (RDSM) model. Model performance was evaluated using 10-fold cross-validation, concordance index (C-index), and overall survival (OS). A multivariable Cox proportional hazards model using the same covariates was constructed as a conventional reference analysis. Survival curves were compared using the log-rank test. Potential confounding was adjusted for using 1:1 propensity score matching (PSM). Patients who underwent LR showed significantly longer OS than those who underwent RFA (5-year survival rate 81.4% vs. 73.1%; P < 0.005). The trained RDSM model achieved a C-index of 0.68, comparable to that of the multivariable Cox proportional hazards model (0.69). In the deep learning (DL) model, patients who received the recommended treatment exhibited significantly longer survival than those who did not (5-year survival rate 81.2% vs. 73.9%; P < 0.005; PSM, 81.9% vs. 76.7%; P < 0.005). The DL model recommended LR in 6,966 (84.5%) patients who received RFA, especially those exhibiting typical imaging patterns (early enhancement and washout in the computed tomography (CT) images [85.9% vs. 61.7% and 84.1 vs.74.3%, respectively]). The proposed DL-based treatment recommendation model may help guide treatment allocation for patients with HCC with up to three nodules measuring ≤ 3 cm. These findings suggest the potential utility of individualized DL-based treatment recommendations.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-17
DOI
https://doi.org/10.1038/s41598-026-78836-0
Primary Topic
Hepatocellular Carcinoma Treatment and Prognosis
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A machine learning algorithm facilitates treatment selection for up to three hepatocellular carcinomas ≤ 3 cm in size

Sumito Yoshida, Ryosuke Tateishi, Ami Kamada, Yoshinari Asaoka et al.
3 links
Scientific Reports
Hepatocellular Carcinoma Treatment and Prognosis
article

A machine learning algorithm facilitates treatment selection for up to three hepatocellular carcinomas ≤ 3 cm in size

Sumito Yoshida, Ryosuke Tateishi, Ami Kamada, Yoshinari Asaoka, Kengo Yoshimitsu, Norihiro Kokudo, Masatoshi Kudo, Naoya Kato, Takumi Fukumoto, Etsuro Hatano, Kiyoshi Hasegawa, Hiroko Iijima, Yoshinori Kabeya, Yasuhide Yamada, Masayuki Kurosaki, Takashi Kokudo, Mitsuo Shimada, Yuma Nakamura, Takamichi Murakami, Hirohisa Yano, Michiie Sakamoto
article en

Abstract

Liver resection (LR) and radiofrequency ablation (RFA) are recommended for patients with up to three hepatocellular carcinomas (HCCs) measuring ≤ 3 cm and preserved liver function. Nevertheless, treatment allocation is decided according to the preference of the treating doctor. This study aimed to develop a predictive model to guide treatment decisions based on survival outcomes. A total of 18,958 patients with HCC with up to three nodules measuring ≤ 3 cm were included from the nationwide survey of Japan. Deep survival analysis was performed using the Recurrent Deep Survival Machines (RDSM) model. Model performance was evaluated using 10-fold cross-validation, concordance index (C-index), and overall survival (OS). A multivariable Cox proportional hazards model using the same covariates was constructed as a conventional reference analysis. Survival curves were compared using the log-rank test. Potential confounding was adjusted for using 1:1 propensity score matching (PSM). Patients who underwent LR showed significantly longer OS than those who underwent RFA (5-year survival rate 81.4% vs. 73.1%; P < 0.005). The trained RDSM model achieved a C-index of 0.68, comparable to that of the multivariable Cox proportional hazards model (0.69). In the deep learning (DL) model, patients who received the recommended treatment exhibited significantly longer survival than those who did not (5-year survival rate 81.2% vs. 73.9%; P < 0.005; PSM, 81.9% vs. 76.7%; P < 0.005). The DL model recommended LR in 6,966 (84.5%) patients who received RFA, especially those exhibiting typical imaging patterns (early enhancement and washout in the computed tomography (CT) images [85.9% vs. 61.7% and 84.1 vs.74.3%, respectively]). The proposed DL-based treatment recommendation model may help guide treatment allocation for patients with HCC with up to three nodules measuring ≤ 3 cm. These findings suggest the potential utility of individualized DL-based treatment recommendations.

Scientific Reports
Kurume University (JP), Hyogo Medical University (JP), Chiba University (JP), Kyoto University (JP), Japan Medical Association (JP), Fukuoka University (JP), National Center for Global Health and Medicine (JP), Tokushima University Hospital (JP), Musashino Red Cross Hospital (JP), IBM Japan, Ltd. (Japan) (JP), Kobe University (JP), The University of Tokyo (JP), International University of Health and Welfare (JP), Hyogo University (JP), Teikyo University (JP), Kindai University (JP)
Good health and well-being
Hepatocellular Carcinoma Treatment and Prognosis
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