Enhancing breast cancer diagnostics: Shape-aware angular feature learning for precision in breast cancer...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Breast cancer diagnosis is critical for survival, but conventional methods struggle in resource-limited settings.
• Differentiating benign from malignant tumors remains a challenge, leading to potential misdiagnosis and unnecessary procedures.

Purpose of the Study:

• To introduce a novel methodology, Shape-Aware Angular Feature Learning (SAAFL), for accurate breast cancer classification.
• To enhance the performance of breast ultrasound (BUS) image analysis, particularly in resource-poor environments.

Main Methods:

• Speckle-Reducing Anisotropic Diffusion (SRAD) filters were used to denoise ultrasound images.
• A robust segmentation approach (RBBSAM-RSF) enabled automatic tumor detection.
• Angular Feature (AF) analysis and a hierarchical classification system combining Support Vector Machines (SVM) and Backpropagation Artificial Neural Networks (BPANN) were employed.

Main Results:

• The SAAFL model achieved 95.38% accuracy on 1293 breast ultrasound images.
• The proposed method outperformed traditional texture-based and morphological models.
• The approach demonstrated reduced false positives and the potential for fewer unnecessary biopsies.

Conclusions:

• SAAFL offers an accurate, noninvasive, and interpretable method for breast cancer diagnostics.
• The model is suitable for near-real-time deployment in low-resource clinical settings.
• Integration of deep learning segmentation with shape-aware feature analysis advances breast cancer detection.