Detection Of Breast Cancer Using Machine Learning On Histopathological Images

Abstract

Breast cancer is still a leading cause of mortality in women all over the world, and an early diagnosis is a key factor in the survival of patients with breast cancer. The focus of this research is the use of ML, more specifically deep learning-based algorithms, to reliably classify Breast Cancer from Histopathological images. Using the publicly available dataset from BreakHis, it has developed a thorough procedure including image pre-processing, image augmentation, and classification of the images based on a variety of classifiers such as SVM, RF, CNN, ResNet50, and InceptionV3. ResNet50 performed the best in all these methods with a classification accuracy of 96.1% and an AUC of 0.98. Deep learning (DL) models, especially those employing transfer learning, performed better than conventional machine learning (ML) methods, because of the ability of deep learning to character features of restricted histological patterns. The study highlights the potential of automated image-based diagnosis to enhance the performance of pathologists, curb diagnostic variation, and boost diagnostic accuracy, particularly in settings with limited resources. Moreover, performance comparison to previous work confirms that the proposed model is robust and universal. The study is promising, although it has several limitations, including the single dataset used and high complexity. It plans to develop XAISC incorporating interpretable AI and multimodality data to promote clinical applications and acceptance in the future.