Noisy Label-Resistant Deep Learning Methods for Cancer Diagnosis in Histopathology Images
Didem Ölçer, Çağatay Berke Erdaş · 2023
Histopathology, one of the anatomical pathology tools, is one of the pathology branches specialising in the histological evaluation of diseased tissues. Histopathological images are examined by pathologists in the diagnosis of cancer and other diseases. Pathologists analyse the general tissue in detail in the diagnostic process, which requires a laborious workload and time. This study proposes a deep learning-based solution for anomaly detection using histopathology images, which is one of the most critical stages in cancer diagnosis. Thus, by developing deep learning techniques to speed up manual evaluations, more effective and close monitoring of sick people is possible. Additionally, the dataset used in this study shows that noisy and mislabeled data are a significant challenge in deep learning models. The accuracy was 0.35 with incorrectly labeled data, and 0.94 with correctly labeled data.