Advancing breast ultrasound diagnostics through hybrid deep learning models

Ajmeera Kiran, Janjhyam Venkata Naga Ramesh, Irfan Sadiq Rahat, Mohammad Aman Ullah Khan, Anwar Hossain, Roise Uddin

Abstract: Highlights•Introduced the EfficientKNN model, a novel hybrid approach combining EfficientNetB3 for advanced feature extraction with k-Nearest Neighbors for robust classification, achieving unprecedented accuracy in breast ultrasound image analysis.•The EfficientKNN model achieved a perfect accuracy rate of 100 %, with precision, recall, and F1-scores all at 100 %, significantly outperforming traditional models such as VGG16, VGG19, AlexNet, and standalone EfficientNetB3.•Employed rigorous image preprocessing and data augmentation techniques, including normalization, resizing, and segmentation with overlay masks, to improve model performance and generalization across varied clinical data.