[Hepatocellular carcinoma segmentation and pathological differentiation degree prediction method based on multi-task...

Area of Science:

• Oncology
• Medical Imaging
• Machine Learning

Context:

• Hepatocellular carcinoma (HCC) is a prevalent liver malignancy requiring accurate segmentation and differentiation prediction for treatment and prognosis.
• Current methods address segmentation and differentiation independently, neglecting task correlations.
• This limitation impacts surgical planning and patient outcome evaluation.

Purpose:

• To develop a unified multi-task learning model for simultaneous HCC segmentation and pathological differentiation prediction.
• To enhance classification accuracy using multi-scale feature fusion and improve segmentation precision with boundary-aware attention.
• To optimize performance across both tasks via a dynamic weighted average multi-task loss function.

Summary:

• A novel multi-task deep learning framework is proposed, integrating segmentation and classification subnets for HCC.
• Key innovations include multi-scale feature fusion for classification and boundary-aware attention for segmentation.
• A dynamic weighted average loss function balances task performance, achieving superior results on a dataset of 295 HCC patients.

Impact:

• The model achieved a Dice similarity coefficient of (83.9 ± 0.88)% for segmentation and an F1 score of (80.05 ± 1.7)% for classification.
• Demonstrates superior performance compared to existing multi-task learning approaches for HCC.
• Provides a valuable tool for clinical diagnosis and treatment strategies for HCC patients.