Grading of hepatocellular carcinoma based on diffusion weighted images with multiple b-values using convolutional...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Hepatocellular carcinoma (HCC) grading is crucial for treatment planning.
• Accurate HCC grading often relies on histopathology, which is invasive.
• Diffusion-weighted imaging (DWI) offers a non-invasive method for HCC assessment.

Purpose of the Study:

• To develop and evaluate a deep learning model for grading HCC using features from multi-b-value DWI.
• To assess the efficacy of fusing deep features from different b-values for improved HCC classification.
• To compare the performance of the proposed method against conventional DWI-based approaches.

Main Methods:

• Retrospective analysis of 100 pathologically confirmed HCC lesions from 98 subjects.
• Logarithmic transformation of DWI images at b-values 0, 100, and 600 s/mm².
• Application of 2D convolutional neural networks (CNNs) for deep feature extraction and fusion, incorporating a deeply supervised loss function.

Main Results:

• The proposed deep feature fusion method achieved 80% accuracy in HCC grading.
• This outperformed direct ADC map feature analysis (72.5%) and single b-value image features (65-70%).
• The area under the curve (AUC) for the proposed method was 0.83, indicating superior performance.

Conclusions:

• Fusion of deep features from multi-b-value DWI using CNNs provides high performance for HCC grading.
• This approach demonstrates potential as a non-invasive tool for HCC lesion characterization.
• Deeply supervised learning enhances the performance of DWI-based HCC assessment.