A Comprehensive Review on Breast Cancer Detection, Classification and Segmentation Using Deep Learning

The incidence and mortality rate of Breast Cancer (BC) are global problems for women, with over 2.1 million new diagnoses each year worldwide. There is no age range, race, or ethnicity threshold, as all women are susceptible; however, no permanent remedy has been developed for it. Therefore, the survival of patients with BC can be improved significantly with an early and accurate diagnosis. There are a number of studies that have created automated approaches employing various types of medical imaging to detect the emergence of BC, but the accuracy of each method varies depending on the resources available, nature of the problem and dataset being employed. However, there is a dearth of review articles that summarize the current research on BC diagnosis. This manuscript addresses the current state of the art in artificial Deep Neural Network (DNN) techniques for BC detection, classification and segmentation using medical imaging. In addition, it emphasizes the working principles, benefits and limitations of imaging modalities used to detect BC, along with a comprehensive analysis of those modalities. The primary purpose of this paper is to identify the most effective imaging modalities and DL approaches that can handle the huge dataset with reliable predictions. The results of this review indicate that mammography and histopathologic images are primarily employed for BC classification. Furthermore, approximately 55% of the research used public datasets while the rest used private data sources. To reduce variability and overfitting in BC images, several studies have used pre-processing methods such as data augmentation, scaling, and normalization. Moreover, distinct neural network architectures, both shallow and deep, are used to analyze BC images. The CNN is widely employed to develop an efficient BC classification model and several studies either used a pre-trained model or created a new DNN. Lastly, this review addressed 13 significant challenges that are encountered throughout the course of the review for future researchers that aim to improve BC diagnosis models using a wide range of imaging techniques. This paper has the potential to be a helpful resource for both beginners and experts in the field of medical image analysis, particularly those who focus on DL based BC detection, classification and segmentation employing a variety of imaging modalities.