Implementing Multilabeling, ADASYN, and ReliefF Techniques for Classification of Breast Cancer Diagnostic through...

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Machine Learning

Background:

• Accurate detection of cancerous areas in low-resolution X-ray images is challenging due to image redundancy and subtle textural variations.
• Existing methods often struggle with classification accuracy when relying on limited or single features.

Purpose of the Study:

• To develop an automated system for multilabel recognition of morphological images and detection of cancerous areas.
• To enhance classification accuracy by combining diverse feature extraction methods and addressing dataset imbalance.

Main Methods:

• Feature extraction using Gray-Level Co-occurrence Matrix (GLCM), Local Binary Patterns (LBP), and other algorithms.
• Combining extracted features and addressing imbalanced data using ADASYN (Adaptive Synthetic Sampling).
• Utilizing the ReliefF algorithm to select significant features and a feedforward neural network for classification.

Main Results:

• The proposed method achieved high performance metrics: 99.5% accuracy (micro and macro), 99.5% precision, 99.5% recall, and 99.4% specificity.
• A low misclassification rate of 0.5% was observed, demonstrating the robustness of the system.
• Validation was performed using the INbreast database.

Conclusions:

• The combined feature extraction and data balancing approach significantly improves the accuracy of cancer detection in X-ray images.
• The developed system shows strong potential for reliable computer-aided diagnosis in medical imaging.