Image-Based Breast Cancer Histopathology Classification and Diagnosis Using Deep Learning Approaches

Abstract

Breast cancer is characterized by abnormal cell growth, which leads to tumor formation. Autonomous breast cancer detection has seen good progress. However, there are still several challenges to robust detection. This survey article explores the complexities inherent in multiclass classification for breast cancer diagnosis, aiming to improve patient care, efficiency, and timeliness. In our study, we focus on using histopathology slide images and assess the current state of breast cancer classification, particularly with artificial intelligence, specifically deep learning and convolutional neural networks. The histopathology images are key tools in the diagnosis process, allowing pathologists to visually assess tissue samples for signs of cancer. Our analysis reveals several challenges that hinder the effectiveness of current diagnostic methods. One significant issue is the need for more diversity in the existing datasets, which often fail to represent a wide range of patient populations. This limitation reduces the accuracy of diagnostic results, mainly when applied to different clinical environments. Furthermore, class imbalances within these datasets, where certain cancer types or stages are underrepresented, lead to biased diagnoses, with more common cases being easily identified while rarer cases are frequently missed. Another challenge is the limited generalizability of current diagnostic techniques, which perform well in controlled environments but often need to improve when applied to new, unseen data from different institutions or imaging systems. Additionally, the complexity of histopathological analysis means that it can be difficult for clinicians to interpret certain findings, leading to uncertainty in the diagnostic process. Our study reveals that addressing these issues requires collaborative efforts to improve the quality of datasets, reduce class imbalances, and develop optimal standardized diagnostic methods. By overcoming these challenges, we can enhance the accuracy, efficiency, and accessibility of breast cancer diagnosis, ultimately leading to better patient outcomes and global healthcare. We believe that by examining several factors and variables and conducting an in-depth analysis of the state of the art, this study will contribute to the state of the art and benefit researchers in both computing and medical domains.