Hounsfield Unit Variations-Based Liver Lesions Detection and Classification Using Deep Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Rising incidence of liver malignancies necessitates advanced diagnostic tools.
• Early detection of liver lesions significantly improves patient survival rates.
• Current diagnostic methods often overlook Hounsfield Unit variations.

Purpose of the Study:

• To develop an automated system for liver lesion detection and classification.
• To integrate Hounsfield Unit analysis into deep learning models for improved accuracy.
• To assist clinicians in diagnosing and treating liver lesions.

Main Methods:

• Utilized deep learning techniques, including Faster R-CNN, R-FCN, SSD, and Mask R-CNN.
• Incorporated Hounsfield Unit density variations from contrast-enhanced and non-contrast CT images.
• Employed a transfer learning approach for model development.

Main Results:

• Achieved up to 97.4% accuracy in detecting and classifying common liver lesions.
• Demonstrated improved performance compared to existing methods across six experimental scenarios.
• Validated the effectiveness of Hounsfield Unit integration for lesion localization and data labeling.

Conclusions:

• The developed deep learning models effectively aid in automatic liver lesion segmentation and classification.
• The system reduces reliance on subjective clinical experience for diagnosis.
• Offers a valuable tool for enhancing the accuracy and efficiency of liver disease management.