February’s Medical Student Grand Rounds at DHMC featured two Geisel School of Medicine students who shared their research on improving patient outcomes through evidence-based quality improvement and a feasibility study of AI-powered at-home smartphone-based ovulation prediction.
Here are summaries of their presentations:
Daniela Armella Tangarife MED’28
Diagnostic Accuracy of Imaging Modalities for Hand Flexor Tendon Injuries: A Systematic Review and Meta-Analysis
“Flexor tendon injuries of the hand present a persistent clinical challenge: physical examination alone is often insufficient, particularly for partial lacerations where pain, swelling, or delayed presentation can obscure findings. While ultrasound (US), MRI, and CT have each been explored as diagnostic adjuncts, their relative performance had never been systematically synthesized—a gap my team and I became intrigued by, given its direct implications for when to image, which modality to choose, and whether exploratory surgery can be avoided. These are questions central to my broader interest in improving patient outcomes through evidence-based quality improvement.
“In collaboration with Stanford Plastic Surgery, I led this systematic review and meta-analysis as first author, searching 12 databases per PRISMA guidelines including PubMed, Embase, Cochrane, and Scopus. Sixteen studies met inclusion criteria—12 examining ultrasound, 6 MRI, and 1 CT—with three studies evaluating both modalities. Random-effects models pooled diagnostic accuracy, with I² statistics for heterogeneity and forest plots generated per modality.
“Our results were interesting. As it turns out, both primary modalities—US and MRI — demonstrated strong performance. Ultrasound achieved pooled sensitivity of 91% and specificity of 97%, while MRI reached 88% sensitivity and 100% specificity. CT showed sensitivity of 92% and specificity of 100%, though based on a single study. These findings support ultrasound as first-line imaging given its lower cost, real-time visualization of tendon movement, and broader clinical availability. MRI remains preferable for high-suspicion cases or complex surgical planning given its superior specificity.
“Limitations of our study include high heterogeneity across studies, variability in US operator experience, and predominantly retrospective designs.
“This work was presented as an e-poster at the American Association of Hand Surgery Annual Meeting in January 2025, and our findings inform a clear call for prospective validation studies, standardized imaging protocols, and cost-effectiveness analyses—steps I see as essential to translating this evidence into meaningful, systemic improvements in surgical care.”
Thao-Mi Nguyen MED’28
Digitally Enabled AI-Interpreted Salivary Ferning–Based Ovulation Prediction: Feasibility Study
“Females with irregular or unpredictable menstrual cycles, including those with polycystic ovary syndrome (PCOS), have limited options for validated at-home ovulation prediction. Most over-the-counter ovulation prediction kits rely on urinary luteinizing hormone (LH) indicators optimized for regular cycles with a predictable mid-cycle LH surge, leaving a significant gap in care. Artificial intelligence (AI) holds potential to address this health deficit through a smartphone-based salivary ferning ovulation test. But first, sufficient training data must be collected from the very population that stands to benefit.
“This project was a collaborative feasibility study by Massachusetts General Hospital, Brigham and Women's Hospital, and the Harvard T.H. Chan School of Public Health, and was recently published in the Journal of Medical Internet Research. The study evaluated whether individuals with irregular menstrual cycles could successfully complete the tasks needed to train a future AI model, including collecting and uploading daily saliva and LH data over two menstrual cycles, attending lab visits, and returning biological samples.
“Of 133 individuals recruited, 69 were eligible, and 43 consented and completed a baseline survey. Of those who received a study kit, 17 began data collection, and 7 ultimately completed the study. Common reasons for withdrawal included cycles being too irregular for the study timeline, pregnancy, relocation, time constraints, and stress related to observing anovulation.
“These findings yielded important insights for future study design, such as employing more targeted recruitment messaging to reduce high ineligibility rates, streamlining procedures to ease participant burden, and incorporating health education to help participants manage the potential emotional impacts of monitoring ovulation. With these optimizations, a scaled version of this study could generate sufficient data to develop and train an AI-powered, smartphone-based ovulation predictor specifically designed for individuals with irregular cycles and PCOS.”
