Machine learning in computational pathology through self-supervised learning and vision transformers

Precision medicine and personalized therapy are less and less synonymous, becoming two fundamental pillars for the proper management of cancer patients. In this scenario, histopathological diagnosis remains a fundamental point for identifying the best possible therapeutic treatment for each patient. In the past few years, the enormous potential of AI in assisting pathologists during daily routine was demonstrated, thanks to the recognition of biological and pathological entities by trained neural networks. Here, we propose a tool to support diagnostic work in a context of shortage of pathologists and increasing number of histological cases by leveraging the most recent and innovative methodology of self-supervised Vision Transformers (ViT): ViTs have a great potential to assist the pathologists in defining the morphological characteristics of tissues and to support them during the diagnostic phase in perfect synergy, as is already the case with Computer Aided Diagnosis in Radiology.