Enhancing Diagnostic Accuracy in Breast Ultrasound Imaging through Deep Learning and Image Augmentation Techniques

This study investigates the application of deep learning models, specifically EfficientNet-B7, combined with advanced image augmentation strategies to improve the diagnostic accuracy of breast ultrasound image classification. The research utilizes a comprehensive dataset of breast ultrasound images labelled as benign, malignant, and normal, supplemented by corresponding masks to highlight areas of interest. By applying image changes such as Random Horizontal Flip, Random Rotation, and Color Jitter, the study intends to address class imbalances and boost the model's generalizability. The EfficientNet-B7 model, fine-tuned on the expanded dataset, attained a classification accuracy of 99% with a precision of 99%, recall of 99%, and F1-score of 99%, demonstrating a considerable improvement in recognizing various circumstances. These metrics underscore the model's effectiveness in minimizing diagnostic errors, crucial for clinical applications. This research highlights the potential of combining high-capacity neural networks with targeted image augmentations to improve outcomes in medical imaging diagnostics.