Effective Tumor Annotation for Automated Diagnosis of Liver Cancer

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Automated tumor annotation is crucial for modern liver cancer diagnosis systems.
• Computer vision and Deep Learning techniques are increasingly used for annotating tumors in biomedical images.
• Accurate tumor annotation provides essential information for clinical reports, including location, size, and characteristics.

Purpose of the Study:

• To propose an effective AI-driven approach for high-quality automated tumor annotation in liver cancer diagnosis.
• To assist radiologists in efficiently generating accurate diagnosis reports by improving tumor segmentation, location, measurement, and recognition.
• To enhance the overall diagnostic process for liver cancer through advanced computational methods.

Main Methods:

• Tumor segmentation using a Multi-Residual Attention Unet to address gradient and diversity issues.
• Liver partition into 8 segments using Multi-SeResUnet for accurate tumor localization.
• Tumor recognition via a multi-labeling classifier based on visual features.
• Tumor size measurement using a proposed regression model.

Main Results:

• The proposed methods demonstrated superior performance compared to state-of-the-art techniques in tumor segmentation, measurement, localization, and recognition.
• Achieved an average tumor size error of 0.432 cm and an annotation accuracy of 91.6%.
• The system shows potential for significantly reducing radiologists' workload and improving diagnostic efficiency.

Conclusions:

• The developed automated tumor annotation system effectively improves the efficiency and accuracy of liver tumor analysis.
• Integration into clinical decision support systems can reduce diagnostic errors and treatment delays.
• The approach has the potential to enhance patient outcomes and streamline clinical workflows.