Deep learning-based breast cancer diagnosis using histopathological images

This The accurate classification of breast cancer histopathological images plays a crucial role in early diagnosis and effective treatment processes. This thesis focuses on the development and performance evaluation of a custom-designed Convolutional Neural Network model for the classification of breast cancer histopathological images. In the initial phase of the research, pre-trained transfer learning models were evaluated on large datasets. Eleven different transfer learning models were utilized, and parameters such as learning rate, loss function, optimization algorithm, and data augmentation techniques were meticulously optimized. Among these models, Xception and InceptionV3 demonstrated the highest performance with accuracy rates of 89,89% and 92.17%, respectively. In contrast, MobileNetV2 and EfficientNetV2B1 exhibited lower performance levels. Based on the results obtained from transfer learning models, a custom CNN model was designed to further enhance classification performance. This model was optimized using squeeze-excite mechanisms and group normalization techniques. The developed CNN model outperformed the transfer learning models with an accuracy of 93,93%, precision of 93,93%, and an F1 score of 93,98%, demonstrating superior performance.