Multi-Task Deep Learning Approach for Simultaneous Objective Response Prediction and Tumor Segmentation in HCC...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Oncology
• Hepatocellular Carcinoma Research

Background:

• Hepatocellular carcinoma (HCC) treatment response assessment and tumor segmentation are crucial for patient management.
• Transarterial chemoembolization (TACE) is a common treatment for HCC, requiring accurate monitoring.
• Current methods for objective response (OR) prediction and tumor segmentation can be labor-intensive and vary in accuracy.

Purpose of the Study:

• To develop and validate a deep learning model for simultaneous objective response (OR) prediction and tumor segmentation in HCC patients post-TACE.
• To evaluate the performance of the proposed multi-task deep learning (multi-DL) model against single-task models and a clinical model.
• To explore the utility of the multi-DL model for patient risk stratification based on predicted treatment outcomes.

Main Methods:

• Development of a multi-task deep learning network (multi-DL) integrating encoder, prediction, and segmentation pathways.
• Retrospective analysis of contrast-enhanced CT images from 248 HCC patients across training, internal validation, and external testing cohorts.
• Comparison of multi-DL with other deep learning methods for OR prediction and tumor segmentation, and with a multivariate logistic regression clinical model for OR prediction.

Main Results:

• The multi-DL model achieved the highest Area Under the Curve (AUC) of 0.871 for OR prediction.
• The multi-DL model demonstrated the highest dice coefficient of 73.6% for tumor segmentation.
• Multi-DL successfully stratified patients into low-risk and high-risk groups with significantly different survival outcomes (p = 0.006).

Conclusions:

• The proposed multi-task deep learning model effectively performs simultaneous OR prediction and tumor segmentation for HCC patients undergoing TACE.
• The integrated approach offers superior performance compared to single-task deep learning models.
• This AI tool shows potential for aiding clinical decision-making in therapeutic regime selection and patient risk stratification.