Publications

Scientific Papers

A Deep Learning Model for Classifying Spontaneous Intracranial Hypotension on Brain MRI — W.F. Wiggins, V. Sahukar, I. Banerjee, M. Lungren, M. Mazurowski, T. Amrhein.
Paper, Radiological Society of North America (RSNA) Annual Meeting, Chicago, USA, Dec 2021.
Paper presented at the Radiological Society of North America (RSNA) Annual Meeting, Chicago, IL, Dec 2021.
We developed a deep learning model using transfer learning with ResNet-50 to identify spontaneous intracranial hypotension (SIH) on brain MRI. Trained on 428 studies from unique subjects, the ensembled model achieved an AUROC of 0.98 across axial and coronal sequences, accurately detecting SIH features such as dural thickening and venous distention. The results demonstrate the potential of AI-assisted diagnosis for improving recognition of underdiagnosed neurological conditions.
Talk Summary PDF: Deep Learning Model for Classifiying SIH on Brain MRI

I’m currently working on a first-author paper in policy steering in robotics. I’ll keep this section updated as work progresses.

Safe Reinforcement Learning in Drone Navigation — Vivek Sahukar, L. Sanneman.
Poster, Arizona State University Southwest Robotics Symposium (SWRS), Tempe, USA, Nov 2024.
This project, developed during the BlueDot Impact Summer School in AI Alignment (UK, Summer 2024), explores how safety constraints can be integrated into reinforcement learning for autonomous drones. Using NVIDIA Isaac Lab, we compared PPO and Constrained Policy Optimization (CPO) in obstacle-rich environments, introducing a danger-zone penalty that encourages safer exploration. Results show that CPO achieves faster, more stable learning with significantly fewer safety violations—demonstrating a practical baseline for safety-aligned robot control.
Poster PDF: Safe RL in Drone Navigation Poster

Coming soon…

For a complete list when it grows, I’ll mirror items on Google Scholar and provide BibTeX.