Deep learning models for predicting the survival of patients with hepatocellular carcinoma based on a surveillance,...

Area of Science:

• Oncology
• Artificial Intelligence
• Biostatistics

Background:

• Hepatocellular carcinoma (HCC) is a prevalent cancer with poor survival rates, necessitating effective long-term follow-up strategies.
• Accurate prediction of patient survival is crucial for managing HCC and guiding treatment decisions.

Purpose of the Study:

• To develop and validate a deep learning model for predicting the survival rates of patients with primary HCC.
• To compare the performance of deep learning models against traditional machine learning and statistical models for HCC prognostication.

Main Methods:

• Utilized data from 2197 HCC patients in the US SEER database (2011-2015).
• Developed and trained deep learning models (DeepSurv, NMTLR) and machine learning models (RSF, CoxPH).
• Assessed model performance using concordance index (C-index), Brier score, ROC curves, and calibration plots.

Main Results:

• Deep learning models, particularly NMTLR (C-index 0.735), outperformed RSF and CoxPH models.
• NMTLR achieved high accuracy in predicting 1, 3, and 5-year survival rates (AUCs 0.824, 0.813, 0.803, respectively).
• The NMTLR model demonstrated superior calibration and discriminative ability for clinical prognostication.

Conclusions:

• The NMTLR deep learning model offers a powerful tool for predicting HCC patient survival.
• This model provides enhanced accuracy and reliability for clinical prognostication and treatment recommendations.
• A web application of the NMTLR model is available for clinical practice, aiding in personalized patient care.