A Hybrid ViT-GRU Model for Breast Cancer Detection: Addressing Class Imbalance Challenges

One crucial use of machine learning in medical diagnostics is the detection of breast cancer, where prompt and precise identification can greatly enhance patient outcomes. Using histopathological images from the Breast Cancer Dataset (400X, 100X, 200X, and 40X magnification levels), we present a reliable deep learning method in this work that uses Vision Transformers (ViT) and Gated Recurrent Unit (GRU) for the categorization of breast cancer. We use a weighted loss function in conjunction with random oversampling approaches to tackle the problem of dataset imbalance, guaranteeing that the model learns from underrepresented classes impartially and efficiently. Outstanding accuracy rates of 98.76% for 400X, 97.84% for 100X, 98.54% for 200X, and 98.63% for 40X photos are attained by our method. These outcomes are supported by outstanding precision, recall, and F1 scores, which show how resilient our model is at all magnification settings. Our ViT-GRU based approach continuously performs better, demonstrating its potency in feature extraction and classification. This study highlights the potential of Vision Transformers in medical image analysis as well as the significance of using creative pre-processing and loss function techniques to address class imbalance. With the potential for increased diagnostic accuracy and dependability, the findings advance automated breast cancer detection.