Artificial intelligence in digital pathology - new tools for diagnosis and precision oncology

Abstract

In the past decade, advances in precision oncology have resulted in an increased demand for predictive assays that enable the selection and stratification of patients for treatment. The enormous divergence of signalling and transcriptional networks mediating the crosstalk between cancer, stromal and immune cells complicates the development of functionally relevant biomarkers based on a single gene or protein. However, the result of these complex processes can be uniquely captured in the morphometric features of stained tissue specimens. The possibility of digitizing whole-slide images of tissue has led to the advent of artificial intelligence (AI) and machine learning tools in digital pathology, which enable mining of subvisual morphometric phenotypes and might, ultimately, improve patient management. In this Perspective, we critically evaluate various AI-based computational approaches for digital pathology, focusing on deep neural networks and 'hand-crafted' feature-based methodologies. We aim to provide a broad framework for incorporating AI and machine learning tools into clinical oncology, with an emphasis on biomarker development. We discuss some of the challenges relating to the use of AI, including the need for well-curated validation datasets, regulatory approval and fair reimbursement strategies. Finally, we present potential future opportunities for precision oncology.

Fig. 1 |. Milestones in computational pathology.

Over the past two decades, technological advances have enabled efficient digitization of whole-slide images, subsequently helping to streamline pathology workflows across pathology labs worldwide. Slide digitization has enabled the creation of large-scale digital-slide libraries, the most popular of which is probably The Cancer Genome Atlas, which has enabled researchers around the world to freely access a richly curated and annotated dataset of pathology images linked with clinical, outcome and genomic information, in turn spurring substantial research activity into artificial intelligence for digital pathology and oncology.

Fig. 2 |. Workflow and general framework for artificial intelligence (Ai) approaches in digital pathology.

Typical steps involved in the use of two popular categories of AI approaches: deep learning and hand-crafted feature engineering.

Fig. 3 |. Visual representations of hand-crafted features across cancer types.

a | Spatial arrangement of clusters of tissue-infiltrating lymphocytes in a non-small-cell lung carcinoma (NSCLC) whole-slide image. b | Features developed using quantitative immunofluorescence of tissue-infiltrating lymphocyte subpopulations (including detection of CD4⁺ and CD8⁺ T cells and CD20⁺ B cells) in NSCLC samples. c | Features reflecting the distribution and entropy of global cell cluster graphs constructed using NSCLC specimens. d | Features computing the relative orientation of the glands present in prostate cancer tissue. e | Diversity of texture of cancer cell nuclei in an oral cavity squamous cell carcinoma. f | Nuclear shape feature computed on cancer cell nuclei in a human papillomavirus-positive oropharyngeal carcinoma. g | Graph feature showing the spatial relationships of different cancer cell nuclei in an oral cavity carcinoma. h | Hand-crafted feature capturing cellular heterogeneity in an oestrogen receptor-positive breast cancer.

Fig. 4 |. Artificial intelligence (Al) and machine learning approaches complement the expertise and support the pathologist and oncologist.

Some of the existing AI approaches currently used by pathologists to analyse images from tumours are depicted. For the practicing oncologist, AI approaches can be used to aid decision making for different aspects of the management of patients with cancer.