Data-efficient medical imaging AI
Learning useful representations from limited, noisy, or unlabeled clinical imaging data.
About
I am a PhD student in the Department of Cyber-Physical Systems at the Indian Institute of Science, Bengaluru, advised by Dr. Punit Rathore. My research focuses on reliable and data-efficient AI for healthcare, with a particular interest in medical imaging, self-supervised learning, representation learning, calibrated risk estimation, and deployment-aware evaluation.
I am interested in learning from challenging clinical data settings where labels are limited, expensive, noisy, or unavailable. My recent work spans self-supervised learning for 3D medical imaging, deterministic autoencoders for structured latent representations, cluster assessment for image datasets, and AI-assisted radiology worklist triage using calibrated risk and queueing simulation.
Before joining IISc, I completed a BS-MS in Electrical Engineering and Computer Science with a minor in Data Science and Engineering at IISER Bhopal.
Research Themes
Self-supervised and representation learning
Designing learning objectives, perturbations, and latent structures for robust medical image understanding.
Reliable healthcare AI
Studying calibration, uncertainty, risk estimation, and evaluation methods for AI systems used in healthcare settings.
Clinical workflow simulation
Translating model scores into workflow-level impact through calibrated risk estimates and queueing simulation.
Publications
-
Two-Fold Patch Perturbation for Efficient Self-Supervised Learning
in 3D Medical Imaging.
Featured
International Joint Conference on Artificial Intelligence and European Conference on Artificial Intelligence (IJCAI-ECAI), 2026.
Paper | Code -
Translating Classifier Scores into Clinical Impact: Calibrated Risk
and Queueing Simulation for AI-Assisted Radiology Worklist Triage.
DAI & AIMedHealth @ AAAI, 2026.
Paper | Code -
DIME: Deterministic Information Maximizing Autoencoder.
DeLTa @ ICLR, 2025.
Paper -
Learning Low-Rank Latent Spaces with Simple Deterministic Autoencoder:
Theoretical and Empirical Insights.
IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2024.
Paper | Project Page | Code -
DeepVAT: A Self-Supervised Technique for Cluster Assessment for Image Datasets.
ViPrior @ ICCV, 2023.
Paper
Selected Projects
Efficient Self-Supervised Learning for 3D Medical Imaging Featured project
Developing patch-perturbation based self-supervised learning methods for 3D medical imaging, with emphasis on label-efficient representation learning.
Keywords: medical imaging, 3D SSL, representation learning, limited labels.
Radiology Worklist Triage Simulation
Translating classifier scores into calibrated risk estimates and simulating their downstream effect on radiology worklist prioritization.
Keywords: calibration, queueing simulation, radiology AI, clinical workflow.
Low-Rank Latent Spaces with Deterministic Autoencoders
Studying deterministic autoencoder architectures that induce structured low-rank latent spaces for classification, clustering, and generative tasks.
Representation Learning for Electronic Health Records
Exploring augmentation strategies and self-supervised time-series representation learning for large-scale EHR datasets.
Datasets: MIMIC-III, PhysioNet 2012.
Collaborators & Mentoring
Dates indicate the period of collaboration or mentorship with me. Affiliations refer to current or most recent affiliations.
Collaborators
Sagar Kumar May 2026 – Present
Work: 3D self-supervised learning for medical imaging.
Mentored Students / Interns
Kabir S. Jamadar Aug 2025 – Dec 2025
Project: Novel perturbation strategy for 3D medical self-supervised learning.
Outcome: IJCAI-ECAI 2026 main conference publication.
Gokul T. Adethya Jun 2024 – Jan 2025
Projects: Medical multi-modal fusion and cross-modal alignment; test-time adaptation for optical flow.
Outcome: Summer internship report.
Internship Report | Project Codebase | Homepage | Google Scholar | GitHub
Teaching
- Teaching Assistant, Probabilistic Machine Learning: Theory and Applications, IISc Bengaluru, Jan–May 2025.
- Teaching Assistant, Machine Learning, IISER Bhopal, Aug–Dec 2022.
- Teaching Assistant, Introduction to Programming, IISER Bhopal, Apr–May 2022.
Service & Technical Leadership
Department GPU Cluster Administrator
- Administer and coordinate access to the department's GPU cluster for research users.
- Support user onboarding, environment setup, resource usage practices, and troubleshooting.
- Help maintain shared compute infrastructure used for machine learning and AI research.
Peer Review
- Reviewer, IJCAI-ECAI 2026.
- Reviewer, DAI & AIMedHealth @ AAAI 2026.
News
- Aug 2026 — Upcoming presentation at IJCAI-ECAI 2026, Bremen, Germany.
- Jan 2026 — Presented work on calibrated risk and queueing simulation for AI-assisted radiology worklist triage at DAI & AIMedHealth @ AAAI 2026, Singapore.
- Apr 2025 — Presented DIME at DeLTa @ ICLR 2025, Singapore.
Gallery
Selected photos from conference presentations, research meetings, and academic events.
AAAI 2026 · Singapore
Blog
Short notes on medical imaging, self-supervised learning, healthcare AI, and research workflows.
Notes on Self-Supervised Learning for Medical Imaging
A short technical note on why self-supervised learning is useful in medical imaging and what makes 3D clinical data challenging.
From Classifier Scores to Clinical Workflow Impact
A note on calibration, risk estimation, and queueing simulation for radiology worklist prioritization.
Beyond Research
Outside research, I have been involved in athletics, martial arts, football, swimming, and music. I served as Coordinator of the Karate Club at IISER Bhopal, won a gold medal in the 4×100m relay at the Inter-IISER Sports Meet, and trained in tabla for five years.
Contact
Email: tirthajitb@iisc.ac.in
GitHub | Google Scholar | ORCID | OpenReview | LinkedIn