Clinical Decision Trees from Clinical Notes * Nilay Kulkarni * MIT Media Lab * February 2025

Abstract:

This talk presents a novel approach to converting unstructured clinical notes into dynamic clinical decision trees, with a specific focus on meningioma treatment decisions. We demonstrate a methodology that processes longitudinal treatment data without requiring semantic factoring, utilizing purely forward updates rather than back propagation. The system employs two distinct medical AI agents - one representing an aggressive treatment approach and another following conservative management protocols - to generate synthetic clinical notes over a 10-year period for virtual patients. These agents make decisions based on comprehensive patient profiles that include demographic information, medical history, and specific meningioma characteristics. The resulting decision trees capture probabilistic treatment pathways that evolve as new cases are introduced, enabling the transfer of clinical decision-making patterns across different healthcare settings. While our approach faces inherent limitations of synthetic data, it offers a promising framework for converting unstructured medical documentation into structured, actionable decision support tools. Our implementation includes a live demonstration platform where users can explore how different patient characteristics influence treatment decisions and observe the dynamic updating of decision probabilities based on accumulated cases.

Speaker Bio:

Nilay Kulkarni is a 24-year-old software developer and entrepreneur who has been developing social impact projects since the age of 14. His work spans multiple domains including public safety, healthcare, and cultural preservation.

His notable projects include a stampede prevention system implemented at the Nashik Kumbh Mela in collaboration with MIT Media Lab, which achieved the festival's first stampede-free year. He has also developed assistive technology for ALS patients and designed an efficient contact tracing system for the Maharashtra government during Covid-19 relief efforts.

Currently, Kulkarni is working on two significant projects. The first involves digital restoration of damaged ancient sculptures in India using generative AI and 3D modeling techniques. The second is the development of Narad, an advanced investment research platform that processes unstructured data using artificial intelligence.