Robust Texture Analysis Using Multi-Resolution Gray-Scale Invariant Features for Breast Sonographic Tumor Diagnosis.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Biomedical Engineering

Background:

• Computer-aided diagnosis (CAD) systems aim to improve diagnostic accuracy and reduce unnecessary biopsies in medical imaging.
• Breast ultrasound analysis is crucial for early detection of breast masses, but its accuracy can be limited by image variability.

Purpose of the Study:

• To develop a robust CAD system for gray-scale breast ultrasound images utilizing texture analysis.
• To evaluate the effectiveness of a novel texture analysis method based on multi-resolution ranklet transform for mass classification.

Main Methods:

• Extraction of gray-scale invariant features using multi-resolution ranklet transform from ultrasound images.
• Application of linear support vector machines (SVMs) on gray-level co-occurrence matrix (GLCM)-based texture features for classification.
• Validation using cross-platform training/testing and leave-one-out cross-validation (LOO-CV) on datasets from three different platforms.

Main Results:

• The proposed ranklet transform-based texture analysis achieved high area under the curve (AUC) values (0.918-0.943) across different databases.
• Performance comparison with wavelet transform showed superior and more consistent results with the ranklet transform method.
• The system demonstrated reduced sensitivity to variations across different sonographic ultrasound platforms.

Conclusions:

• Texture analysis using multi-resolution gray-scale invariant features via ranklet transform is effective for developing a robust CAD system for breast ultrasound.
• The proposed method offers improved diagnostic performance and robustness compared to traditional wavelet transform-based approaches.
• This approach has the potential to enhance the reliability of computer-aided diagnosis in breast mass evaluation.