An Advanced Hybrid Deep Neural Network for Precise Multiclass Classification and Detection of Cancerous and Lymphatic Node States

Metastatic cancer, defined by the spread of cancer cells from their original site to distant body regions, presents considerable diagnostic hurdles. Precise identification in histopathological images is essential for optimal patient care. This research introduces an innovative hybrid framework integrating ResNet50, self-attention modules, and Gated Recurrent Units (GRUs) to enhance binary classification precision in metastatic cancer detection. In opposite to CNN-GRU, CNN-LSTM, and AlexNet-GRU models, our approach showcased remarkable performance across two datasets. For the Histopathologic Cancer Detection BreakHis dataset, the model achieved 96.00% accuracy, 95.32% precision, 96.15% sensitivity, and 95.32% specificity. On the BACH dataset, it attained 98.00% accuracy, 98.44% precision, 98.44% sensitivity, and 98.44% specificity. In multi-class classification tasks, the model achieved an impeccable score of 1.00 on both the BreakHis and BACH datasets. Those outcomes highlight the model's capacity to significantly decrease diagnostic errors and improve pathologists' diagnostic accuracy surpassing other approaches in the field.