Abhishek Desai, MD

Research

Autonomous capabilities in surgical robotics are no longer prospective. Who makes sure they're safe before they reach the operating room?

I study emerging surgical technologies, with a current focus on autonomous surgical robotics and AI governance in surgery. My earlier work used computer vision and natural language processing to augment surgical risk prediction with multimodal unstructured data.

Autonomous surgical robotics

I lead the Autonomous Surgery Working Group within the Robotics Committee of SAGES, where we're working to guide the development of safe and effective autonomous surgical systems. Our current project is SURGE (Surgical Understanding, Reasoning, Generation, and Execution), a taxonomy for classifying autonomous capabilities of surgical robotic systems, intended to provide a shared vocabulary that surgeons, engineers, and regulators can use to evaluate these systems before they reach patients.

In my 2025 chief resident grand rounds at Rutgers, Show Me the Money: How I Learned the Bitter Lesson and You Will Too, I examined the tension between Rich Sutton's bitter lesson (general methods that leverage computation eventually win) and John Searle's Chinese room (a program can appear to understand without understanding) as they apply to surgery. The first portion of the talk is adapted as an essay, Show Me the Money, Part 1.

Abdominal wall reconstruction

From 2021 to 2022, I was the Clinical Research Fellow at the Penn Center for Human Appearance, where I studied machine learning methods to augment surgical risk modeling with unstructured data, with a focus on abdominal wall reconstruction and incisional hernia.

We used computer vision to identify morphometric features on preoperative CT scans that were highly predictive of incisional hernia, validated these features as biomarkers, and developed a morphometry-based model of how incisional hernias form.

We also built a natural language processing pipeline to label key details in operative notes (eg: the type of suture used to reapproximate fascia). This let our group pull specific intraoperative details from large unstructured datasets into our risk models without extensive manual chart review.

I also co-authored a textbook chapter on minimally invasive anterior component separation and helped secure more than $500,000 in grant funding, including funding for two projects I initiated that have continued to grow since I left the lab.

Early research

During medical school, I was a research specialist in the Eskandar Lab at Massachusetts General Hospital. I overhauled the design of the OpBox, an open-source operant conditioning environment for small rodents. We built ten working units and cut the per-unit cost by 35%, bringing the final design to less than 10% of the cost of commercial systems at the time. I also analyzed rapid-sample EEG to study neural activity during decision making and wrote diagnostic software to debug EEG signal anomalies.

Selected publications

The full list is on Google Scholar and in my CV. If you're working on related problems, I would love to connect and collaborate.