Leveraging EfficientNet-B3 with Advanced Fine-Tuning for Precise Breast Cancer Classification
Geomol George, Shanmugam Anusuya · 2024
This study explores the use of the pre-trained EfficientNet-B3 model for classifying breast cancer images into benign, malignant, and normal categories. The model was finetuned with layer-specific adjustments, advanced regularization techniques, and comprehensive layer fine-tuning to enhance its ability to detect complex patterns. The results demonstrated exceptional performance, with benign cases achieving a precision of 98%, recall of 94%, and F1 score of 96%; malignant cases with precision of 90%, recall of 96%, and F1 score of 93%; and normal cases showing precision of 94%, recall of 96%, and F1 score of 95%. The model also achieved 99.84% training accuracy, 0.2964 training loss, and 95.25% test accuracy with a loss of 0.4333. These findings underscore the potential of this advanced EfficientNet-B3-based method to improve diagnostic accuracy and support timely, reliable treatment decisions in breast cancer care.