Multi-scale textural feature extraction and particle swarm optimization based model selection for false positive...

Area of Science:

• Medical Imaging and Artificial Intelligence
• Biomedical Engineering
• Radiology

Background:

• Current mammography Computer Aided Detection (CAD) systems suffer from high false positive rates, leading to avoidable breast biopsies.
• Reducing false positives is crucial for both mass and calcification detection CAD systems in clinical use.

Purpose of the Study:

• To address the problem of false positive reduction in mammography.
• To improve the detection of lesions and masses by optimizing CAD system performance.

Main Methods:

• Analysis of breast tissue using multi-scale textural descriptors (wavelet and gray-level co-occurrence matrix).
• Application of Particle Swarm Optimization (PSO) for feature selection and Support Vector Machine (SVM) classifier optimization.
• Evaluation using datasets from Digital Database for Screening Mammography (DDSM) and Mammographic Image Analysis Society (mini-MIAS).

Main Results:

• PSO-based model selection effectively optimized classifier hyperparameters and parameters.
• Proposed textural features, particularly those from co-occurrence matrices of wavelet representations, showed promising performance.
• Demonstrated efficiency in reducing false positives in mammogram analysis.

Conclusions:

• The proposed method using PSO for model selection and wavelet-based textural features significantly improves mammography CAD system accuracy.
• This approach offers a promising solution for reducing false positives and unnecessary biopsies in breast cancer screening.