Next Generation: Two Backgrounds, One Purpose
Next Generation is a Brigham Clinical & Research News column penned by students, residents, fellows and postdocs. If you are a Brigham trainee interested in contributing a column, email us. This month’s column is written by Tiffany Chen, MD, and Jana Lipkova, PhD, who are postdoctoral fellows in the Mahmood lab.
We are two postdoctoral research fellows in the Artificial Intelligence (AI) for Pathology lab led by Faisal Mahmood, PhD, at the Brigham. As co-authors of a recently published paper in Nature Medicine about using artificial intelligence to predict cardiac transplant rejection, we would like to share our experiences as two individuals who came together from different backgrounds to create tools for pathologists.
Tiffany: During my second year of pathology residency, I noticed that the field of pathology was rapidly changing. A digital revolution was coming, and I wanted to be at the forefront. During the Discover Brigham event in 2019, I came upon a poster from Dr. Mahmood’s lab that showcased a novel machine learning algorithm for breast cancer diagnosis. I saw how the fusion of new techniques like machine learning — which uses data and experience to automatically improve analysis — and histopathology — which relies on the expertise of pathologists to assess tissue to make a diagnosis — could change the way pathologists practiced. As much as I wanted to get involved, I was worried about my lack of computing knowledge given that I only knew very basic coding. I spent time looking up things like, “What is a neural network?” and completing Coursera courses before I finally had the courage to join Dr. Mahmood’s lab. When I joined the lab, I was even more nervous that I would stick out like a sore thumb. However, I soon discovered that everyone was there for the same purpose — to build out tools that will help patients — and we each had unique skills and perspectives to bring to the team. I was partnered together with Jana to tackle a diagnostic problem within cardiac pathology: cardiac transplant rejection. While I collected all the clinical cases, Jana worked on developing the algorithm that could serve as a tool for pathologists.
Jana: I did my PhD in computer science in radiology, so joining a pathology lab was quite a jump into new water. I was attracted by the possibilities of examining diseases at the tissue and cellular levels and the potential of computational pathology on not only impacting patient diagnostics, but also treatment. I was stunned by the beautiful pathology images and had no idea how to interpret them. After joining the lab, I watched several YouTube videos to get some sense of the data, but I quickly realized how difficult the task is, as each tissue and disease have their own manifestations. I secretly admired Tiffany’s understanding of the field. We were both looking at the same picture, but each of us was seeing something else: me, mainly the pink and purple colors, and Tiffany, the clinically relevant insights. For our cardiac project, Tiffany explained to us the whole patient journey, the biopsy assessment process, and its impact on the patient treatment. Thanks to Tiffany’s insights, we designed a model addressing all the major diagnostic tasks for rejection detection, which is not a very common practice in computer science. Technical people often study clinical problems in isolation or under very specific conditions which differ from clinical practice. This probably also contributes to the skepticism for AI methods among pathologists. However, by working together with Tiffany, we broadened the scope and the potential of our model, which could lead to easier downstream adoption in the clinical setting.
Joining forces
When there is a new field like computational pathology that combines two views (computer science and pathology), clinical and technical partnerships are a must. From the technical side, having a clinical partner on board was a fantastic experience. It was so liberating to have someone with clinical expertise to fill in the gaps in our medical understanding. The same could be said for the clinical side as well, since we were good at identifying issues, but lacked the technical expertise to truly solve them. Working together made us both more confident in our work and helped us build tools that were computationally sound and clinically relevant and feasible. For our cardiac project, we worked efficiently and synergistically together while avoiding obstacles may have arisen from either clinical or technical knowledge gaps. Our success in our project was only possible thanks to the unique way that Dr. Mahmood built an interdisciplinary lab- fostering a collaborative and open space for people from different scientific backgrounds to freely interact, learn and explore.