CEUS-based classification of liver tumors with deep canonical correlation analysis and multi-kernel learning

Area of Science:

• Medical Imaging
• Oncology
• Artificial Intelligence

Background:

• Contrast-enhanced ultrasound (CEUS) is crucial for diagnosing liver cancers.
• Radiologists analyze typical enhancement patterns across arterial, portal venous, and late phases for diagnosis.

Purpose of the Study:

• To develop a computer-aided diagnosis (CAD) system for liver cancer detection.
• To simulate the clinical diagnosis process using only three typical CEUS images from different phases.

Main Methods:

• Utilized deep canonical correlation analysis (DCCA) for multi-view fusion of CEUS image pairs (arterial-portal venous, arterial-late phases).
• Employed a multiple kernel learning (MKL) classifier with six-view features generated by DCCA.
• Developed a CEUS-based CAD system mimicking radiologist analysis.

Main Results:

• The proposed DCCA-MKL algorithm demonstrated superior performance in distinguishing benign liver tumors from malignant liver cancers.
• The CAD system effectively integrates multi-phase CEUS information for enhanced diagnostic accuracy.

Conclusions:

• The DCCA-MKL approach offers a reliable and effective method for computer-aided diagnosis of liver cancers using CEUS.
• This technique simulates clinical practice, potentially improving diagnostic efficiency and accuracy in liver tumor assessment.