Automated ultrasonography of hepatocellular carcinoma using discrete wavelet transform based deep-learning neural...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Hepatocellular carcinoma (HCC) diagnosis relies heavily on medical imaging.
• Abdominal ultrasound (US) is a common, non-invasive imaging modality for liver assessment.
• Accurate and early detection of HCC is crucial for patient outcomes.

Purpose of the Study:

• To introduce HCC-Net, a novel wavelet-based artificial neural network for HCC diagnosis from US images.
• To develop a hierarchical lesion detection and pattern-augmented classification system.
• To improve the accuracy and reliability of automated HCC detection and classification.

Main Methods:

• Utilized discrete wavelet transform (DWT) for US image decomposition.
• Implemented a hierarchical coarse-to-fine lesion detection framework (CoarseNet and FineNet).
• Employed a pattern-augmented classifier using synthesized wavelet components for enhanced feature distinction.

Main Results:

• Achieved high diagnostic performance on the Samsung Medical Center dataset.
• Reported an accuracy of 96.2%, sensitivity of 97.6%, and specificity of 98.1%.
• Demonstrated the effectiveness of the coarse-to-fine detection and pattern-augmented classification approach.

Conclusions:

• HCC-Net shows significant potential as a reliable tool for liver cancer screening.
• The proposed method enhances lesion detection and classification accuracy in abdominal US images.
• Wavelet-based image decomposition and pattern augmentation improve diagnostic capabilities.