Liver Tissue Classification Using an Auto-context-based Deep Neural Network with a Multi-phase Training Framework.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Hepatocellular carcinoma (HCC) diagnosis relies on accurate liver tissue classification.
• Acquiring 3D annotated segmentation masks for magnetic resonance imaging (MRI) is costly and time-consuming, leading to small training datasets.
• Existing neural network approaches struggle with limited data for precise liver tissue classification in HCC.

Purpose of the Study:

• To develop and evaluate a novel deep convolutional neural network (CNN) for classifying liver tissue types in 3D multi-parameter MRI.
• To address the challenge of small annotated datasets in HCC imaging by incorporating auto-context elements into a U-net-like architecture.
• To improve the accuracy and robustness of liver tissue classification in patients with hepatocellular carcinoma.

Main Methods:

• Designed a novel deep convolutional neural network (CNN) integrating auto-context into a U-net-like architecture.
• Employed a patch-based strategy with weighted sampling for effective training on limited data.
• Implemented a multi-resolution and multi-phase training framework to enhance model regularization and reduce the learning space.

Main Results:

• The proposed method demonstrated promising results in classifying liver tissue on 3D multi-parameter MRI.
• The novel deep learning approach outperformed standard neural network methods.
• The model also showed superior performance compared to a benchmark method for liver tissue classification in HCC patients.

Conclusions:

• The developed deep learning model effectively classifies liver tissue in 3D MRI for hepatocellular carcinoma patients.
• The novel architecture and training framework successfully address the challenge of limited annotated data.
• This approach offers a significant advancement for automated liver tissue classification in oncology imaging.