Leveraging EfficientNetB3 for Accurate Breast Cancer Classification: Insights and Innovations
Parul Nasra, Sheifali Gupta, Gotte Ranjith Kumar · 2024
One of the main causes of death among women globally, breast cancer emphasises the great need of a correct and fast diagnosis. Mammography and histology are among the conventional diagnostic methods that mostly depend on professional interpretation and so are prone to human mistake. This work improves the accuracy and efficiency of breast cancer classification using EfficientNetB3, a state-of- the-art convolutionial neural network. EfficientNetB3 is well-known for its capacity to scale depth, width, and resolution, therefore enabling enhanced feature extraction free from major computational demand. We use the Breast Ultrasound Images dataset in this work, fine-tuning the model to categorise images into benign, malignant, and normal categories by including many image augmentation strategies. EfficientNetB3 yields a classification accuracy of 94% coupled with great precision, recall, and F1-scores across all categories according to experimental results. Accuracy and loss curves as well as a thorough confusion matrix study help to confirm the model's strong generalising capacity. This work not only shows the promise of EfficientNetB3 in medical diagnostics but also offers understanding of how flexible the model is to fit several imaging modalities. Reducing misclassification rates helps to ensure more accurate and effective diagnosis systems, especially between benign and malignant categories. This study opens the path for implementing artificial intelligence-driven solutions in clinical environments, therefore aiding early identification and better patient outcomes in breast cancer treatment.