Tumor Detection In Breast Histopathological Images Using Faster R-CNN

Breast cancer is the most common type of cancer in women, and it is crucial to detect it at an early stage for a better prognosis. There are various ways of detecting and confirming breast cancer that are highly dependent on the imaging modalities, such as mammograms, ultrasound, magnetic resonance imaging (MRI), and histopathological image analysis by pathologists. With the help of recent progress in machine vision and artificial intelligence, computational methods for image analysis such as deep learning have been widely applied for automated decision-making in breast cancer diagnosis using feature extraction and localization. This study utilizes Faster Region-based Convolutional Neural Network (Faster R-CNN), one of the deep learning algorithms for tumor detection in annotated breast histopathological images, and analyzes the effects of two pre-processing procedures (color normalization and patching) on the images for optimization of the Faster R-CNN model. It was observed that the model’s sensitivity drastically increased from 1 % to 60 % by patching the images. The effect of image color normalization was conditional and improved results for only a few cases.