Transfer Learning with DenseNet for Accurate Breast Histopathology Images Classification

Breast cancer is one of the most common diseases in women globally, and early detection is essential for effective treatment and improved patient outcomes. In this work, a transfer learning method using pretrained convolutional neural networks (CNNs) in MATLAB is presented for the automatic categorization of breast histopathology pictures. The DenseNet architecture is specifically utilized due to its enhanced feature propagation and efficient parameter utilization, both of these are being demonstrated to be beneficial in medical imaging applications. The suggested approach makes use of a labeled dataset of histopathological pictures that have been scaled and preprocessed to meet the network's input specifications. In order to achieve binary classification between benign and malignant tissue samples, the pretrained DenseNet model's final layers are fine-tuned using transfer learning. Accuracy, confusion matrix, and visual classification results are employed to measure the model's performance after it is trained using MATLAB's Deep Learning Toolbox. Experimental results indicate the transfer learning technique uses the rich and dense feature representations acquired from large-scale picture datasets to achieve excellent classification accuracy even with a very small dataset. To further improve accuracy and generalizability in practical diagnostic contexts, future research will investigate patch-based categorization, ensemble models, and bigger datasets.