Machine learning, deep learning, and artificial intelligence as applied to the field of cytopathology: a...

Imaging informatics approaches have played a significant role in gynecologic cytology since the introduction of BD FocalPoint and the ThinPrep Imaging System at the dawn of the new millennium. Although rudimentary compared to modern artificial intelligence and machine learning (AI/ML) techniques, the development of these systems nearly 20 years ago led to increased diagnostic performance, with potential to address issues relating to speed, throughput and reliability of cytologic assessment. However, rather than ushering in an age of increasing automation and quantitative analysis, algorithmic advances in cytology (and pathology in general), stagnated for approximately a decade after its introduction. In the 2010's, the advent of AlexNet, coupled with plummeting storage and GPU costs, open-source development environments and the marketing of affordable, high throughput digital slide scanners accelerated progress and interest in digital pathology AI/ML applications. While the widespread adoption of large-scale consortium level datasets (e.g., The Cancer Genome Atlas) spurred the development of AI/ML methods for surgical pathology, large-scale collection, and algorithmic modeling of cytology specimens has lagged behind due to the varied nature of cytology preparations and assessment. However, increasing attention has been paid to cytopathology with a commensurate increase in publications and commercial applications in this area. Still, the applications of AI/ML to cytopathology and the herculean effort to implement fully digital pathology services remain in its nascent stages. This review explores the current state of research and commercial development in digital cytopathology, with a focus on AI/ML technologies.