AJAP2201 - CME/CMLE - The utility of unsupervised machine learning in Anatomic Pathology-The utility of unsupervised machine learning in Anatomic Pathology

The utility of unsupervised machine learning in Anatomic Pathology

Pdf Summary

The review article discusses the utility of unsupervised machine learning techniques in anatomic pathology due to the lack of annotated datasets. The document introduces unsupervised learning methods, including clustering, generative adversarial networks (GANs), and autoencoders, and explains how these techniques can address the shortage of labeled data in pathology. Clustering is discussed as a method for grouping similar instances of data together and its application in semisupervised learning. GANs are explored for generating synthetic data and performing color normalization, while autoencoders are highlighted for unsupervised pretraining to transfer learned representations to classifiers. The application of these techniques in pathology, including stain normalization, transfer, and synthetic data generation for algorithm training, is detailed. The challenges and potential of using GANs and autoencoders in pathology are explained, emphasizing their role in improving supervised learning algorithms by addressing the lack of labeled datasets. Additionally, the review provides insights into the technical aspects, challenges, and potential applications of these unsupervised techniques in developing artificial intelligence in pathology.

Keywords

unsupervised machine learning

anatomic pathology

clustering

generative adversarial networks

GANs

autoencoders

labeled datasets

stain normalization

algorithm training

artificial intelligence