Magnification Prior: A Self-Supervised Method for Learning Representations on Breast Cancer Histopathological Images

This work presents a novel self-supervised pre-training method to learn\nefficient representations without labels on histopathology medical images\nutilizing magnification factors. Other state-of-theart works mainly focus on\nfully supervised learning approaches that rely heavily on human annotations.\nHowever, the scarcity of labeled and unlabeled data is a long-standing\nchallenge in histopathology. Currently, representation learning without labels\nremains unexplored for the histopathology domain. The proposed method,\nMagnification Prior Contrastive Similarity (MPCS), enables self-supervised\nlearning of representations without labels on small-scale breast cancer dataset\nBreakHis by exploiting magnification factor, inductive transfer, and reducing\nhuman prior. The proposed method matches fully supervised learning\nstate-of-the-art performance in malignancy classification when only 20% of\nlabels are used in fine-tuning and outperform previous works in fully\nsupervised learning settings. It formulates a hypothesis and provides empirical\nevidence to support that reducing human-prior leads to efficient representation\nlearning in self-supervision. The implementation of this work is available\nonline on GitHub -\nhttps://github.com/prakashchhipa/Magnification-Prior-Self-Supervised-Method\n