A NOVEL DEEP LEARNING APPROACHES FOR MULTI-CLASS HISTOPATHOLOGICAL SUB-IMAGE CLASSIFICATION USING PRIOR KNOWLEDGE

Early diagnosis of breast cancer is critical for effective treatment and reducing mortality rates. Computer-aided diagnosis tools have become essential for identifying and diagnosing cancer in its initial stages. Convolutional neural networks (CNNs) have shown significant promise in medical image analysis, aiding in the detection of cancer cells and the classification of histopathological images through advanced data processing techniques. This study introduces a novel framework that combines transfer learning (TL) with an Incorporation of Prior Knowledge algorithm for multi-class classification of breast cancer using histopathological images. A new dataset comprising 3,600 sub-image histopathological images is presented, generated from the original Bach dataset. The study evaluates various pre-trained deep neural networks, including Inception V3, VGG19, GoogleNet, ResNet 101, and NASNet. Notably, the integration of prior knowledge and the focus on sub-image classification rather than whole images significantly enhanced cancer classification accuracy. The proposed method, leveraging the NASNet architecture, achieved a remarkable classification accuracy of 98.61%. Additionally, this study advances beyond conventional classification tasks by investigating tumor localization within breast cancer, utilizing sub-image analysis to improve diagnostic precision and support effective clinical decision-making. This innovative approach enhances classification performance and contributes to more accurate tumor localization, thereby significantly improving diagnostic capabilities in breast cancer detection