This piece, by Chhavi Chauhan, PhD, and Rama Gullapalli, PhD, first appeared on the Digital Pathology Association (DPA) blog and has been cross-posted here with permission.
We are finally witnessing increasing adoption of digital pathology in clinical practice. Pathologists are beginning to embrace and utilize artificial intelligence (AI) tools and ideas to augment routine pathology practice. All stakeholders in pathology need to fully understand the ethical frameworks required to develop, deploy, and maintain transparent, trustworthy, explainable, scalable, interoperable, and sustainable digital AI tools.
We had the privilege and honor of partnering up for Digital Pathology Association (DPA)’s recent webinar on the topic of “AI Ethics: Healthcare Lessons to Guide Pathology Practice” in which we highlighted ethical concerns pertaining to AI initiatives in the healthcare domain followed by a discussion on how pathologists can leverage these ethical AI frameworks within the practice of pathology to augment and improve it for the betterment of humankind.
While there is indeed a lot of enthusiasm for the use of pathology originated data, there are a lot of concerns among the pathologists regarding the appropriate use of this data. The specialty of pathology is well over 150 years old. Over this time, pathologists have carefully put together key elements of the specialty to serve the patients in an ethical manner. This includes diagnostic algorithms using histology, classification of diseases, and implementing technology to measure small analytes which make a huge impact on a patient. Nearly 70% of the data that goes into enabling a patient diagnosis and/or clinical management is believed to originate from pathologists and the lab. It therefore becomes imperative that a newcomer technology like AI and Machine Learning be put through its paces in an ethical and rigorous manner, to ensure that there are no untoward effects on patients.
Some of the interesting applications of AI are in the use of prediction of molecular status (e.g., microsatellite instability, BRAF mutational status, etc.) based on histomorphological image data, and have immense implications. There are some hidden patterns within histology, apparent to AI algorithms, that human eyes cannot yet discern. By learning and understanding these in more detail, pathologists can become better practitioners of the craft of pathology in the future.
We are excited about the potential of AI to enable “multi-modal” pathology in the future. Multi-modal pathology which will entail the integration of multiple, lab-based data elements of a patient (e.g., histomorphology, molecular data, biochemical and hematological data) into providing much more robust predictors of a patient’s disease course. The multi-modal paradigm is already present to some degree within the practice of hematopathology which makes use of assays such as histomorphology, next-generation sequencing assays, flow cytometry indices, and cytogenetics into enabling diagnoses and also predictions of disease outcomes. Hopefully, there will be such clinical practice paradigms within solid-tumor oncology as well. Many practicing pathologists focus to a great degree on oncology alone and tend to neglect non-oncology clinical pathology, which may be a mistake. There is huge potential for the adoption and use of AI-enabled clinical pathology practice for non-oncology related diseases (e.g., cardiovascular disease, gastrointestinal diseases and autoimmune rheumatology). We neglect these areas at our own peril!
A common complaint among pathologists interested in understanding and participating more in AI-enabled pathology is that they are often relegated to tasks such as circling slides as a “gold standard” and do nothing else. Though AI/ML engineers do possess advanced knowledge of AI algorithm development, development of many of these algorithms are critically dependent on the domain knowledge that a pathologist possesses. Cross-domain skills and knowledge (e.g. AP and CP knowledge) are critical to ensure robust development of AI algorithms. Additionally, as purveyors of the fundamentals of disease pathogenesis, pathologists are in a unique position to bring to table the basic knowledge of disease pathogenesis necessary to enable complex, multi-modal AI algorithm development. A practical skill set that we often take for granted, i.e., quality improvement and quality control in pathology, is going to be critical to ensure the ethical development of AI algorithms in future pathology practice. Learning data-analytic skills in the domain of pathology informatics is going to be immensely useful in the future as well. A pathologist who can read, review, and work with large datasets will be in much demand in the future.
Finally, ethics in healthcare often gives the mistaken impression of being focused on the topic of moral ethics alone. This could be the farthest from the truth! The field of AI ethics is a vibrant and dynamic emerging specialty. Experts from a wide-range of specialties (e.g., engineers, computer scientists, philosophers, lawyers, law makers, policy experts) are beginning to come together to enable ethical AI. A big missing puzzle thus far has been the voice of the end-users (e.g., pathologists themselves). There is a huge-need for the domain experts to lend their expertise/voice in the development of ethical AI within each specialty. Pathologists have a huge opportunity here to enable AI development in a manner that considers many of these issues.
We discussed many of these topics within our webinar that generated an engaging discussion, especially on the use of AI-based tools as well as the need for incorporation of a dedicated course on “Ethical AI in Digital Pathology” in various pathology curricula.
Sadly, the emerging topic of AI ethics is too vast and extensive to cover comprehensively within the time limits of a single webinar. However, pathologists are invited to review many of the key ethical issues that are likely to impact the practice of pathology in the coming decades. Who knows? there is perhaps an opportunity to develop a niche of ethics expertise that you could yourself contribute to!
Pathologists are key stakeholders in developing and deploying sustainable and ethical AI in digital pathology. Their educated efforts and guidance can help us avoid the mistakes that have emerged with time in the healthcare domain as we embark on our journeys to make digital pathology mainstream. Pathologists can and must play a key role in the development of AI-enabled digital technologies based on sound principles of ethical AI.
To view the webinar recording, visit DPA’s webinar archive.
The upcoming 2022 PRIM&R Annual Conference (PRIMR22) will bring the entire research community together and serve as a forum for discussing common themes that span the research enterprise, including ethical considerations for AI and data. Attendees will have the opportunity to network and address challenges facing research programs, their administrators, and leaders. PRIMR22 will take place online December 12-15, with in-depth workshops on December 6 and December 9.
Chhavi Chauhan, PhD, is Director of Scientific Outreach, American Society for Investigative Pathology and Director of Continuing Medical Education Program for the Journal of Molecular Diagnostics. She is also a Co-Chair of the DPA Membership Committee.
Rama Gullapalli, PhD is Assistant Professor, Department of Pathology and Assistant Professor, Department of Chemical and Biological Engineering at the University of New Mexico.