Mitosis Detection In Breast Histopathology Image Using Ensemble Features Fed CNN Model

Breast cancer is acknowledged as the second leading cause of women’s cancer-related deaths. Quantifying mitosis in pathology sections can provide invaluable insight into both the prognosis and tumor aggressiveness. The task of detecting mitotic cells in histopathology images, which is crucial for predicting the grade and aggressiveness of breast cancer, poses a demanding and time-consuming challenge for pathologists. In this work, we establish an automated approach for detecting breast cancer based on the identification of mitotic counts from histopathology images. In CNN approaches, the generation of features is predominantly unsupervised, and handcrafted features are aimed at modeling domain-relevant attributes. Consequently, a hybrid feature dataset is constructed by incorporating handcrafted features. This dataset can be reorganized and utilized as input for a CNN architecture to classify feature values as either mitotic or non-mitotic. Subsequently, various data combinations are employed to train the CNN model for classification, and the most accurate feature set is chosen to optimize prediction accuracy. In our study, the shape feature has an F-score of 0.87.