"I met patients with larynx and thyroid cancers," Dalal explained. "This allowed me to understand the actual data we were dealing with and how an AI-guided diagnosis or prognosis could truly help in a clinical setting."
"While joining me in clinic, he was able to see how the treatment of head and neck cancer can impact a person's ability to communicate and additionally what side effects patients can experience from the treatment," Goyal said, explaining this firsthand experience allowed the team to refine their focus and look for ways AI could identify exactly which patients would benefit from extra treatments and which could safely avoid them.
A 'smart assistant' for doctors
The research team focused on using computers to analyze medical images, like CT scans. By training AI models, they aimed to create a clinical decision support tool — essentially a "smart assistant" to help doctors make better, faster decisions for their patients.
In addition to Goyal, other team members included Daniel Asante Otchere, a research computing software engineer focused on AI and machine learning (AI/ML) on the Research Innovations with Scientists and Engineers team at the Penn State Institute for Computational and Data Sciences (ICDS) hired through Penn State CTSI to support its Informatics Core, and Christopher Tseng, PGY-3 Resident in Otolaryngology. Under this team’s mentorship, Dalal contributed to the team's research on laryngeal cancer, while also gaining valuable exposure to their broader work on thyroid imaging. The team's major projects include:
- Predicting laryngeal cancer recurrence: After patients have surgery for laryngeal — voice box — cancer, doctors need to know if the cancer is likely to come back. The team built a deep learning model that looks at tissue images and patient data to predict this risk. The model identified which high-risk patients would benefit from follow-up radiation.
- Automating thyroid scan segmentation: When patients get CT scans, it can be very time-consuming for doctors to manually measure the thyroid gland and any lumps, or nodules, on it. The Penn State research team developed a 3D AI model to map out the thyroid and any nodules with excellent accuracy of the gland.
- Classifying "accidental" thyroid nodules: Often, patients get a CT scan for one reason, and doctors incidentally find a nodule on their thyroid. The Penn State researchers used an AI model to help determine the significance of these and whether they are cancer.
The power of collaboration and mentorship
Big ideas need a starting point, and none of this would have been possible without pilot funding from the Penn State CTSI, according to the researchers. The research team received a seed grant from the CTSI that they said provided the time and resources to take on this complex informatics project — and to bring Dalal on board.
Despite his age, Dalal said he was never intimidated by the advanced team.
"Daniel was very open," Dalal recalled. "He said that, as a young researcher, I can sometimes bring fresh ideas for models to try, and he kindly gave me the freedom to explore different approaches. The whole team was really kind and trusted me."
That trust paid off. Otchere said he was blown away by Dalal’s abilities.
"I was genuinely surprised by how quickly Shaun grasped complex concepts and methods, even topics that typically take much longer to internalize," Otchere said. He noted that Dalal could take advanced ideas — like using "multi-instance learning" to train AI when data is limited — and apply them with very little help. "That ability to work independently and confidently at such an advanced level was both impressive and impactful for the team."
But building a smart AI isn't just about writing good code — it has to actually work in a real hospital. To ensure this, Goyal and Tseng met regularly with Dalal to review his models.
For Tseng, an otolaryngology resident who double-majored in biology and computer science, mentoring Dalal was especially meaningful because he said he recognized his own career path in the young student.
"Being both a clinician and a programmer comes with its rewards and challenges," Tseng said, explaining that doctors and engineers often have very different ways of solving problems. "Being able to bridge this gap in perception, to facilitate productive collaboration between physicians and engineers ... is a skill unto itself, one that Shaun has consistently performed at a high level."
This bridging of skills was crucial during their regular check-ins, Goyal said.
"The collaboration between data scientists and clinician researchers allows investigators to explore tools that can directly shift clinical care," Goyal said. "It is also very rewarding to be able to mentor and work with intelligent and focused students like Shaun, who clearly represent the future of clinical research and clinical care."
Real-world impact: Looking to the future
The research team said they envision this technology changing real lives. According to Otchere, after further testing and validation, these AI models could be integrated into hospitals to provide clinicians with better diagnostic and prognostic tools. Furthermore, Otchere said, this technology could potentially help reduce diagnostic errors in low-resource settings, like busy emergency rooms or rural hospitals.
As for Dalal, his work at Penn State has set him on a bright path. His research helped him win his local science fair and earn college scholarships. He also presented his work at the Regeneron International Science and Engineering Fair, and he will present at the International American Head and Neck Society Conference in Boston this summer.
This fall, Dalal is heading to Duke University to study computational biology, with the long-term goal of earning both a M.D. and a Ph.D. to become a physician-scientist.