Dr. Aristotelis Tsirigos is an Associate Professor of Pathology and Director of the Applied Bioinformatics Laboratories at the NYU School of Medicine. He received his Ph.D. in Computer Science from Courant Institute of New York University in January 2006, and his B.S. from the National Technical University of Athens, Greece in 1998. Dr. Tsirigos has more than 15 years of experience in genomics and machine learning at NYU and IBM Research. His research focus is primarily in cancer epigenetics and AI-based diagnostics for cancer, cardiovascular and other diseases.
Lab website: http://www.tsirigos.com/
Dr. Tsirigos, congratulations on your recent paper in Nature Medicine [24(10):1559-1567]. Could you explain in layman’s terms what is the issue you are addressing in this study, the main findings, and what is next?
In our recently published work, we studied whether it would be possible to automate and streamline certain aspects of cancer diagnosis. As we all know, cancer is a devastating disease that occurs when cells in our body stop functioning normally. Unfortunately, this can happen in many different ways causing several different types of cancer whose precise diagnosis is critical for deciding the appropriate treatment. Diagnosis is performed by expert doctors (pathologists) – with several years of training – who use a microscope to look at samples obtained from patients’ tumors. Oftentimes, this process takes several minutes and is prone to human error and bias. What we showed is that we can “teach” an artificial intelligence (AI) algorithm to look at the same images seen by the doctors and make an automated and accurate diagnosis within seconds. AI algorithms are capable of automatically associating features on an image (for example, shape and density of cancer cells) to specific cancer types after being shown enough examples. In our study, we used more than 1,000 images to teach the AI algorithm to distinguish two common types of lung cancer. These findings can have a direct impact on patient care: faster, more accurate and more accessible diagnosis for cancer patients.
Do you see a future where Artificial Intelligence and Deep-learning software could replace a healthcare practitioner?
Over the past few years, AI has made great progress on several challenging tasks, ranging from speech recognition and translation to driving autonomous vehicles. In principle, AI can facilitate some aspects of a healthcare practitioner’s tasks, thereby reducing human error. It can also help doctors evaluate all available clinical data more efficiently while allowing more time for patient care. I do not think that doctors and AI compete for the same job, AI will be another valuable tool for the doctors.
What other projects are you working on?
After focusing on lung cancer as a proof of principle, we are now expanding and testing our models on other types of cancer and soon on different types of disease. In addition, we are planning to integrate different data that may be available for each patient, such as MRI images, lab tests, and genetic information. Ultimately, going beyond an accurate diagnosis, what is of greatest value for the patient is to choose the most appropriate treatment and realize the vision of “Precision Medicine”. AI promises to attack this problem decisively, but there is still a long way to go.
What was a turning point or defining moment in your work as a scientist?
I believe that you become a real scientist through hard work. But the defining moment comes when you realize that you have a vision, a vision that you relentlessly pursue with passion and a lot of patience. A defining moment for me was after we had published our first high-impact papers. My close collaborator and I were discussing future projects, basically asking ourselves “and now, what is next?”. Our answer to this question was that high-impact papers are not enough, that’s not the goal, because unfortunately, “we haven’t cured cancer yet”. To do that, we, as scientists, need to stay focused on the goal and constantly reinvent ourselves.
If not a scientist, what else would you rather be?
I love thinking of new questions, looking for answers, questioning myself, challenging the current knowledge and redefining what is possible. Therefore, it is very hard to imagine a more appropriate profession!
Tell us what you like to do when you aren’t working on research
What I like the most is spending time with my 7-year old son: it is amazing what I can learn from him, kids are a pure source of inspiration! Other than that, I try to keep up with my hobbies: volleyball and classical guitar.