The Pathologist and Artificial Intelligence Applications in Histopathology

By: Douglas P. Malinowski, Ph.D. VP of Clinical Affairs

Artificial intelligence, and the application of computer-based learning to solve complex problems, has become a major advancement in recent years. When artificial intelligence is applied to human healthcare to help in complex diagnosis and clinical management problems, the clinical utility, accuracy, validity, and reproducibility of the AI output are of paramount importance. To ensure safety and clinical efficacy in these AI applications, the FDA has provided guidance for the development and subsequent analytical and clinical validation of these AI applications. These guidance documents provide a regulatory framework to support regulatory approval of such artificial intelligence devices for use in the medical field.  To date, there are over 500 medical devices or software applications as medical devices which have been reviewed and cleared by the FDA for implementation in patient health. The vast majority of these approvals are in the field of radiology, which as a field of medicine has been quick to adopt new technologies related to telemedicine and artificial intelligence.

Histopathology and morphology assessment of biopsy specimens are a cornerstone of diagnosis for human disease including cancer. Yet, this pathology assessment is critically dependent upon the experience and interpretative skills of the pathologist. In cancer, the impact of human interpretation, and associated inter-observer variability in the final diagnosis is well documented [1]. For example, in breast cancer, the pathology assessment of histologic grade provides important information regarding the state of tumor differentiation and degree of proliferation. And both of these features are important in providing a prognostic assessment of breast cancer and the likelihood of cancer growth and recurrence [2], yet breast cancer histologic grade has documented variability [3].

With this limitation in mind, histopathology diagnosis has been supplemented to additional information, including hormone receptor status using IHC assays and the use of molecular markers and genomic signatures which offer a more  perceived level of objectivity and also provide more detail about the molecular characteristics of the cancer which are not easily determined based upon standard morphology assessment. However, application of AI to cancer pathology has the capability of detecting morphology features, either undetected or unrecognized during histology assessment, which provides a more objective, quantitative assessment of tumor types and provides additional detailed information about tumor types, characteristics and behavior that extends the utility of standard morphology assessment of histopathology.

The applications of artificial intelligence algorithms to help the pathologist in the review and final diagnosis of a cancer pathology slide have been documented recently in the literature. These include identifying suspicious regions for potential cancer, the concept of re-screening pathology slides from cases which have been signed out a non-disease, to identify cancer which was overlooked during the initial assessment. This has the ability to reduce false negative rates in cancer detection. Other applications include the detection and assessment of histology features which are arduous, time consuming or subject to variable interpretation (histologic grade in breast cancer; Gleason score in prostate cancer; enumeration and quantitation of mitotically active cells in a tumor; detection of metastatic cancer cells in lymph node biopsies; and cancer prognosis and risk of recurrence). For example, in two recent publications, AI was shown to accurately assess breast cancer histologic grade in comparison to expert pathologist review [4] and AI was shown to improve breast cancer prognosis and risk stratification for recurrence relative to standard clinical features [5].

Furthermore, recent FDA approvals for the use of digital pathology, and associated software applications, to aid in pathologist review of cancer pathology slides represent a step forward in terms of clinical utility and broader adoption in the marketplace.

With continued use and application, the implementation of AI assessment of cancer histopathology specimens is expected to enhance and extend the utility and information on cancer biology leading to greater insights into the prediction or risk of recurrence, proliferation status, degree of tumor differentiation, etc.

Rather than being viewed as a replacement to standard pathology, the integration of AI into standard histopathology assessment is poised to enhance and extend the pathologist’s interpretation and ultimate diagnostic characterization of a patient’s biopsy specimen leading to more accurate characterization of disease status and highlighting treatment options for the patient.

 

References:

[1] Elmore JG, Longton GM, Carney PA, Geller BM, Onega T, Tosteson AN, Nelson HD, Pepe MS, Allison KH, Schnitt SJ, O’Malley FP, Weaver DL. Diagnostic concordance among pathologists interpreting breast biopsy specimens. JAMA. 2015 Mar 17;313(11):1122-32. doi: 10.1001/jama.2015.1405. PMID: 25781441; PMCID: PMC4516388.  https://pubmed.ncbi.nlm.nih.gov/25781441/

[2] Schwartz AM, Henson DE, Chen D, Rajamarthandan S. Histologic grade remains a prognostic factor for breast cancer regardless of the number of positive lymph nodes and tumor size: a study of 161 708 cases of breast cancer from the SEER Program. Arch Pathol Lab Med. 2014 Aug;138(8):1048-52. doi: 10.5858/arpa.2013-0435-OA. PMID: 25076293.

[3] Ginter PS, Idress R, D’Alfonso TM, Fineberg S, Jaffer S, Sattar AK, Chagpar A, Wilson P, Harigopal M. Histologic grading of breast carcinoma: a multi-institution study of interobserver variation using virtual microscopy. Mod Pathol. 2021 Apr;34(4):701-709. doi: 10.1038/s41379-020-00698-2. Epub 2020 Oct 19. PMID: 33077923; PMCID: PMC7987728. https://pubmed.ncbi.nlm.nih.gov/33077923/

[4] Mantrala S, Ginter PS, Mitkari A, Joshi S, Prabhala H, Ramachandra V, Kini L, Idress R, D’Alfonso TM, Fineberg S, Jaffer S, Sattar AK, Chagpar AB, Wilson P, Singh K, Harigopal M, Koka D. Concordance in Breast Cancer Grading by Artificial Intelligence on Whole Slide Images Compares With a Multi-Institutional