AI-DRIVEN Novel Approach for Liver Cancer Screening and Prediction Using Cascaded Fully Convolutional Neural Network.

Area of Science:

• Medical Imaging
• Computational Biology
• Artificial Intelligence in Medicine

Background:

• Automated segmentation of liver lesions is crucial for biomarker analysis but is challenging due to variations in lesion characteristics and imaging parameters.
• Current methods struggle to accurately determine liver cancer stages based on lesion patterns, facing obstacles in training accuracy.
• Accurate liver segmentation is essential for reducing error rates in diagnostic algorithms.

Purpose of the Study:

• To develop an automated system for detecting liver tumors and lesions in abdominal MRI scans.
• To improve the accuracy and efficiency of liver cancer diagnosis through advanced image analysis techniques.
• To address challenges in lesion segmentation and staging using novel 3D shape parameterization and deep learning.

Main Methods:

• Utilized 3D affine invariant and shape parameterization for lesion detection and modeling.
• Employed geodesic active contour analysis for initial liver segmentation.
• Implemented Cascaded Fully Convolutional Neural Networks (CFCNs) for tumor area segmentation and error rate minimization.

Main Results:

• Achieved 94.21% accuracy for liver tumor analysis using CFCNs.
• Demonstrated a total accuracy rate of 93.85% across training and testing datasets.
• Completed analysis with a calculation time of less than 90 seconds per volume.

Conclusions:

• The proposed system effectively automates the detection of liver tumors and lesions in MRI.
• The combination of 3D shape analysis and CFCNs significantly enhances diagnostic accuracy.
• This automated approach offers a promising tool for liver cancer research and clinical applications.