“Trust your gut” is a well-worn adage. But when it comes to how we look at gut health, Tanmay Shukla believes doctors might do well to say, “Trust AI.” 

In 2023, Shukla graduated from the Quantitative Biomedical Sciences (QBS) program (now the Health Sciences MS programs) at the Geisel School of Medicine at Dartmouth. Today, he works at Geisel, using machine learning algorithms to analyze radiology and histopathology images—an approach that is often better and faster than the human eye.  

Shukla has quickly become an expert in helping clinicians accurately detect, diagnose, and treat inflammatory bowel diseases (IBD). “Inflammatory bowel disease is one of the most prominent in the United States,” he explains. “Right now, everyone is talking about their gut health. Because the gut is the source of your health, it could be the source of disease.” 

Before coming to Hanover, Shukla’s peregrinations—from India to England, Singapore, and Brazil—may resemble that of a fickle globetrotter. But upon closer inspection, it’s clear that his travels helped him gain a suite of technical chops whose versatility has brought him to where he is now.  

Shukla earned his engineering degree in his home country of India, then traveled to London to study machine learning and statistics, working primarily with finance data. At the National University of Singapore, he transitioned away from finance into deep learning of medical imaging and genomics. “Working in the medical field was more promising,” he reflects. “I could make an impact by developing a technique that could assist in the detection of cancer or some disease that’s hard to diagnose, which would be overall better for society.” 

After a brief stint as an AI engineer in Brazil, Shukla landed at Geisel. During the QBS program, he met Saeed Hassanpour, PhD, who serves as the director for the Center for Precision Health & Artificial Intelligence (CPHAI) and professor of biomedical data science, computer science, and epidemiology.  

“The best part of the health data science at QBS is being taught by the professors who work closely with doctors at [Dartmouth Hitchcock Medical Center (DHMC)],” Shukla says. “If you’re a doctor interested in doing something with AI, you can contact a professor who’s an expert in machine learning. The professors work with students. And for students, it’s a big opportunity to learn. So, it’s a perfect fit for doctors, professors, and students. Everything’s all in one place.” 

After Shukla came to Dartmouth, he took classes in epidemiology and health informatics. Around this time, he took classes with Hassanpour, who took Shukla under his wing and taught him about “proper medical imaging techniques.” Under Hassanpour’s tutelage, Shukla grew his knowledge of deep learning, biostatistics, neural networks, and probabilistic modeling methods in the analysis of medical images. Today, he serves as Technical Program Manager at CPHAI, spending much of his time at the Hassanpour Lab, which develops new tools to analyze biomedical data and advance precision medicine.  

“Tanmay’s research exemplifies the power of interdisciplinary collaboration between AI experts and clinicians in advancing precision medicine,” Hassanpour says. “His work on developing algorithms to analyze medical images is particularly impactful for improving the diagnosis and treatment of inflammatory bowel diseases.” 

In the Hassanpour Lab, Shukla has teamed up with Amit Das, a senior Dartmouth undergraduate student majoring in computer science. Together, they set out to build an algorithm to accurately differentiate between different types of IBD—such as Crohn’s disease, colitis, or an indeterminate cause—a critical first step for pathologists to make an accurate diagnosis. 

After accurately classifying IBD types, Shukla and Das refined their model to precisely determine disease activity, a crucial clinical need. Shukla explains, “Doctors want to know if IBD is ‘active’—causing symptoms like pain, diarrhea, and bleeding due to gastrointestinal inflammation—or ‘inactive,’ where the disease is in remission without symptoms.”  

Now, the AI model can reliably distinguish active from inactive IBD. This fact alone accelerates disease detection. But Shukla has not stopped there. Currently, he is developing a four-level classification system for histopathology slide images based on whether the IBD is inactive, mild, moderate, or severe. “These categories help pathologists decide how they’ll treat the disease,” Shukla explains. 

The research project has not been without bottlenecks and snags. One major hurdle was detecting tiny neutrophils, a type of white blood cell that measures between 9-15 micrometers, in extremely high-resolution images, which can strain computational bandwidth. “One of the biggest problems with histopathology images is that they are huge—millions and millions of pixels,” Shukla says. While tumors are on the tissue level of detail and can be easily spotted in an image, neutrophils are much, much smaller. “Our pipeline wasn’t designed for cellular-level detail, so we had to solve that first.”  

Shukla says Hassanpour has been an invaluable resource in learning how to troubleshoot such complex issues. “Whenever we hit a roadblock, Dr. Hassanpour provides invaluable guidance, offering strategic solutions and alternatives to explore.’ And when we need some expert opinion, he’ll connect us with radiologists, pathologists, doctors who can help us more and look at our results,” Shukla says.  

“By leveraging machine learning, Tanmay is helping to create tools that enhance diagnostic accuracy and provide more personalized treatment plans, ultimately leading to better patient outcomes. His dedication and innovative approach are paving the way for significant advancements in the field of precision medicine,” Hassanpour says. 

As Shukla continues to refine his IBD classification system and visualization techniques, his research holds immense promise for improving patient outcomes and streamlining diagnostic processes. Guided by mentors like Hassanpour and with the support of Dartmouth’s interdisciplinary ecosystem, Shukla is poised to develop a pioneering clinical tool that could transform therapeutic practices.  

“I don't think AI will ever replace doctors,” he says. “But it will probably replace doctors who don't use it.” 

Written by: Jeremy Martin