Ultrasound classification of breast masses using a comprehensive Nakagami imaging and machine learning framework

Area of Science:

• Medical Imaging
• Ultrasound Technology
• Biomedical Engineering

Background:

• Breast lesion classification relies on accurate imaging techniques.
• Ultrasound B-mode scans offer non-invasive diagnostic potential.
• Parametric imaging using Nakagami distribution can enhance tissue characterization.

Purpose of the Study:

• To investigate Nakagami parametric images for non-invasive breast lesion classification.
• To determine optimal window parameters for generating Nakagami images.
• To evaluate the efficacy of extracted features for breast cancer diagnosis.

Main Methods:

• Generated seven types of Nakagami parametric images using a sliding window technique.
• Analyzed four window sizes, including column windows (0.1875, 0.45, 0.75 mm) and a standard square window.
• Extracted 72 morphometric, elemental, and hybrid features for classification.
• Performed feature selection to identify the most discriminative features for breast cancer classification.

Main Results:

• A column window of 0.75 mm yielded the highest classification accuracy (93.08%) and Area Under the ROC Curve (AUC) of 0.9712.
• The optimal feature set achieved a 0% False Negative Rate (FNR) and an 8.65% False Positive Rate (FPR).
• Comprehensive analysis of Nakagami parametric images for breast lesion classification was performed.

Conclusions:

• Nakagami parametric images generated with a 0.75 mm column window are effective for breast lesion characterization.
• The proposed method shows potential for accurate and reliable breast cancer diagnosis.
• This approach may help reduce false positive diagnoses in clinical practice.