Performance Evaluation of Hybrid Meta-Heuristic Optimization Techniques in Multilayer Perceptron-Based Breast Cancer Detection

Abstract

Breast cancer is a prevalent malignancy in women worldwide. Mammography screening is an effective method for early detection, and timely diagnosis is essential for improving treatment outcomes and survival rates. Medical imaging techniques, especially mammography, are frequently used for detecting breast cancer and monitoring tumour progression. However, this diagnostic technique has a low positive predictive value for breast biopsies, often resulting in unnecessary procedures for abnormal findings. The Breast Imaging Reporting and Data System (BI-RADS) was developed as a standardized diagnostic technique for reporting mammographic results. It categorizes abnormal findings related to breast cancer into specific groups, facilitating the assessment of breast biopsies and the identification of tumours. In this research, ensemble models are integrated with optimization techniques to enhance the detection process in mammography. The initial step involves pre-processing using a Gaussian Bilateral Filter (GBF) to reduce noise and improve image quality. Following this, the region of interest (ROI) is identified employing the Adaptive Histogram Thresholding and Contour Clustering (AHT-CC) algorithm. Feature extraction is conducted utilizing the Convolutional VGG-16 (ConV-16) model. The extracted features are then fed into a Hybrid Bayesian Genetic Optimized Multilayer Perceptron (HBGOMLP-EnML) for precise identification of breast cancer based on BI-RADS categories. The analysis reveals that the proposed model achieves an accuracy of 98.6%, precision of 97.1%, recall of 97.2%, and F1-score of 97.2% across different classifiers. The results indicate that the proposed method is competitive with existing classification models in the literature, presenting a practical and beneficial solution.