A deep learning-based psi CT network effectively predicts early recurrence after hepatectomy in HCC patients

Area of Science:

• Radiology
• Oncology
• Artificial Intelligence

Background:

• Hepatocellular carcinoma (HCC) has a high recurrence rate, impacting patient prognosis.
• Early prediction of HCC recurrence is crucial for timely intervention and improved outcomes.

Purpose of the Study:

• To develop and validate a deep learning (DL) model for predicting early recurrence in HCC patients.
• To leverage enhanced CT scans and an attention mechanism for improved prediction accuracy.

Main Methods:

• A DenseNet-based neural network was developed using multi-institutional triphasic enhanced CT scans.
• An attention mechanism was integrated to focus on survival-relevant regions.
• Model performance was evaluated using concordance index (C-index), calibration, and decision curve analysis.

Main Results:

• The DL model achieved an AUC of 0.797 for two-year outcomes in the validation cohort.
• The best model demonstrated a C-index of 0.774, with cross-validation yielding an average C-index of 0.778.
• Class activation maps (CAMs) indicated the model focused on relevant intra-abdominal organs.

Conclusions:

• The developed DL-based enhanced CT network shows significant potential for predicting early HCC recurrence.
• This approach represents a promising new strategy for early recurrence prediction in HCC management.