Deep learning for image-based liver analysis - A comprehensive review focusing on malignant lesions

Area of Science:

• Medical image analysis
• Artificial intelligence in radiology
• Deep learning for oncology

Background:

• Deep learning, especially convolutional neural networks (CNNs) and fully convolutional networks (FCNs), is prevalent in medical image analysis.
• Focal liver lesions, including hepatocellular carcinoma and metastatic cancer, require accurate analysis.
• Liver parenchyma and vascular structures are critical for diagnosis and treatment planning.

Purpose of the Study:

• To review deep learning methodologies for analyzing liver structures and lesions.
• To focus on segmentation, object detection, and classification tasks for liver imaging.
• To identify high-performing deep learning approaches based on a systematic literature review.

Main Methods:

• A qualitative search identified 91 relevant papers from journals and conferences.
• Papers were categorized into eight groups based on methodology.
• Performance was evaluated using Dice scores for segmentation and accuracy for classification/detection.

Main Results:

• Hybrid models achieved superior performance in liver and lesion segmentation.
• Ensemble classifiers demonstrated better results for vessel segmentation.
• Combined approaches excelled in both lesion classification and detection tasks.

Conclusions:

• Deep learning methods offer powerful tools for liver image analysis.
• Specific architectures and hybrid approaches yield optimal results for different tasks.
• This review provides insights into effective deep learning strategies for liver lesion and structure analysis.