The Early Adopter

How did your career begin?

Although I had always wanted to pursue a career in pathology, I found it to be the hardest subject at medical school. I presumed I would find something easier – but, after rotating through every other discipline, the spark had caught fire and I set my sights on becoming a pathologist. When I immigrated to the US from South Africa during my training, I noticed the stark difference in the levels of computerization and electronic records between the two countries. Despite great eagerness to use artificial intelligence (AI) in South Africa today, a lack of resources and vendors prevents computational pathology from taking hold. The distinct lack of pathologists – all of whom have high-volume workloads – makes it the perfect setting for AI to make a real difference, but they just don’t have the capability to deploy it.

When I arrived in the US, I was immediately attracted to pathology informatics and, thanks to the mentorship of Bruce Beckwith, I was never afraid to try new technologies. Once I arrived at the University of Pittsburgh Medical Center, I had the perfect Petri dish in which to think big and grow the clinical potential of digital pathology and AI.

How will AI’s impact be felt in the future?

There will be positives and negatives for everybody. On the one hand, I believe that AI will excite people and get them even more interested in their daily work. For trainees, it will be extremely intriguing to work with the technology that will shape the future of their field. I’ve involved trainees in the validation of several AI algorithms and many have subsequently written to me to express their enthusiasm. But there are fears that, because AI has the potential to take over much of the heavy lifting in pathology, it will leave people de-skilled.

At this stage, so few labs have experienced routine AI use that we just don’t know its true impact yet. As we traverse this new and exciting landscape, it’s important to be cautious and validate new algorithms thoroughly, ensuring that we keep a close eye on them. We also shouldn’t make the assumption that AI will always lead to improvement. In a trial of a new algorithm that counts mitotic figures in breast cancer, we asked 27 users – ranging from second-year trainees to senior pathologists – to use the AI tool. Although the algorithm improved efficiency overall, five of the users were slower with AI and we have yet to understand why. We are heading into the unknown and there will inevitably be some failed deployments alongside the many success stories.

How can institutions gain from being early adopters of AI?

I’m a strong supporter of the partnership between academia and industry. In the drug world, the partnership between pharmaceutical companies and academic institutions results in a win-win-win situation: a win for the pharma industry, a win for the academic institution (by advancing science, improving their reputation, and increasing publication rate), and a win for the field in general. I see the same thing happening in pathology with the introduction of AI.

Early adopters will have the opportunity to work with AI start-ups and co-develop deep learning algorithms, whether by providing ideas for applications or supplying data for development. For algorithms that have already been built, institutions can provide the platform to integrate them into routine practice by testing whether they fit into existing workflows. But there are also indirect benefits to adopting AI. The reputation of institutions that adopt the technology is enhanced and noticed by patients, colleagues, and trainees alike – and trainees are increasingly attracted to institutions that can offer a digital pathology and AI platform.

Most excitingly for me, I can now use AI to make new discoveries about disease, which is exactly why I became a pathologist. For example, we have an algorithm that can automatically screen urine cytology slides to make a diagnosis. Using it, I can now analyze close to two million cells and search for new disease patterns or correlations, which wouldn’t have been possible manually. It’s not all about performing mundane tasks when it comes to AI; rather, once these tools are at your disposal, you can start to run quality assurance on data and images while carrying out routine work. It’s like having a new microscope in the lab!

What is your greatest personal achievement?

I founded the Journal of Pathology Informatics and, with the help of forward thinking colleagues including Anil Parwani and Mike Becich, the publication is now 10 years old, of which I am immensely proud. It was affiliated with the Association for Pathology Informatics (API) and provided a vehicle through which everyone involved in the field could share their research, whether or not it was successful. I think it has helped to shape the field of pathology informatics by bringing together pathologists, computer scientists, and clinicians to focus on a central theme. Consequently, I’m thrilled to have been selected for the API’s 2019 Lifetime Achievement Award.

What are your biggest interests outside the lab?

I love music! I especially enjoy going to rock concerts and watching Broadway musicals, the latter being a common family outing. Unfortunately, I don’t get enough time to play golf or read, but maybe AI will facilitate that!