Pixels to Prognosis: ResNet50 Hyper-parameter Analysis for Predicting Benign vs. Malignant Breast Cancer from Biopsy Scans

Breast cancer diagnosis is a critical task that can benefit from advancements in deep learning and computer vision. This study focuses on the analysis and application of the ResNet50 architecture to classify benign and malignant breast cancer types using RCL biopsy images. Three versions of the ResNet50 model were trained and evaluated using diverse hyperparameter configurations, optimization strategies, and learning rate schedulers to determine the most accurate and robust classifier. The performance of each model was assessed using metrics such as precision, recall, F1-score, and accuracy, with additional validation through k-fold cross-validation. ResNet50 (Classifier 3), utilizing the AdamW optimizer and ReduceLROnPlateau scheduler, emerged as the top performer with an average accuracy of 90% and balanced performance across all metrics. These findings underscore the potential of ResNet50, when optimized effectively, for reliable breast cancer classification. The research highlights the importance of model tuning and evaluation in developing robust AI-driven diagnostic tools.