Research · AI · Healthcare Asmit Ganguly I build AI that explains itself.
Biomedical Engineering PhD student at USC building interpretable medical AI for MRI, neuroimaging, and clinically useful decision support.
Curiosity, translated into useful systems.
Personal Moments
A glimpse into my journey and experiences
Medical AI, built for clinical trust.
I’m Asmit — a Biomedical Engineering PhD student at the University of Southern California and a researcher in the Laboratory of Functional MRI Technology (LOFT).
My current work focuses on interpretable deep learning for MRI and neuroimaging, including arterial spin labeling methods for the early detection and classification of Alzheimer’s disease.
I came to medical AI through computer science, earning my BTech from IIT Patna before moving into biomedical engineering. That path still shapes how I work: rigorous about systems, but always anchored to the clinical question.
The goal is simple to state and difficult to earn: models that clinicians can interrogate, trust, and use responsibly.
Core Expertise
Selected publications
Work focused on making complex models more interpretable, accessible, and useful in real settings.
Cracking the code: enhancing interpretability and accessibility of ophthalmic disorders detection using KAN
SPIE Photonics West Ophthalmic Technologies XXXV
Innovative approach to ophthalmic disorder detection using Kolmogorov-Arnold Networks (KAN) for enhanced interpretability.
SEMANTIFY: Unveiling Memes with Robust Interpretability beyond Input Attribution
IJCAI 2024 Main
Advanced meme interpretation using robust interpretability techniques beyond traditional input attribution methods.
AlpaPICO: Extraction of PICO Frames from Clinical Trial Documents Using LLMs
Method
Novel approach for extracting PICO (Population, Intervention, Comparison, Outcome) frameworks from clinical trial documents using Large Language Models.
Quantum Adaptive Sensing for Accelerated MRI
arXiv preprint · 2026
Adaptive k-space sampling formulated as a QUBO to improve accelerated MRI reconstruction at 20% and 10% sampling.
Things I’ve built
A mix of research prototypes, full-stack products, and experiments across software and hardware.
TalkitOut
Full-fledged open-source social media platform with chat, video call, and blogging capabilities. Features real-time communication, community posting, and secure file transfer with AI Co-Pilot integration.
Domain Specific Question Answering (DevRev)
Gold medalist solution for Inter IIT 11.0. Domain adaptation in open domain QA using theme-specific rankers and novel resource allocation algorithms for paragraph examination.
Neural Activity Classifier
Deep learning model for EEG data classification to determine if specific actions are undertaken or thought by patients. Achieves 75% accuracy in single channel, applicable to gaming and VR enhancement.
Ping Pong on FPGA
Standalone ping pong game implemented on FPGA Board DE1-SOC. Integrated hardware design with Linux installation, demonstrating computer architecture principles in embedded systems.
LostFound Android App
Android application for Lost and Found system using Firebase for data storage. Features one-click item return, picture support, and mail verification for enhanced security.
Assembler Emulator
Custom assembler and emulator for a specially designed instruction set. Emulates a 2-register computer with data segment storage using C++, demonstrating low-level programming concepts.
Notes from the in-between
Small ideas, images, half-finished thoughts, and things that don’t need to become a research paper.