Salman Siddique Khan

I am a Research Scientist at Rice University and The Bookout Center for Medical Innovation, Houston Methodist Academic Institute. My research spans machine learning, computational imaging, and clinical AI, with a current focus on developing intelligent systems for cardiovascular care, surgical planning, and disease severity assessment. I finished my Ph.D. from the Computational Imaging Lab, IIT Madras, where I worked under Dr. Kaushik Mitra on computational photography and lensless imaging. As a Ph.D. candidate, I was awarded the Qualcomm Innovation Fellowship (2020–21) and the IUPRAI Best Doctoral Dissertation Award (2024). Prior to my Ph.D., I completed my undergraduate studies in Electronics and Instrumentation Engineering from the National Institute of Technology Rourkela, India.

Email  /  Google Scholar  /  Github  /  LinkedIn

I work at the intersection of machine learning, computational imaging, and clinical AI. My recent research focuses on developing end-to-end AI systems for digital health — including cardiovascular imaging, surgical planning, and disease severity assessment — as well as novel lensless camera systems with applications in gaze tracking, face recognition, and privacy-preserving imaging.

We propose a weakly-supervised framework for coronary artery calcium (CAC) segmentation from non-gated CT scans, using only volume-level CAC scores as supervision — enabling risk stratification without ECG gating.

We introduce a large-scale lensless face recognition dataset and benchmark enabling real-world evaluation of lensless imaging systems.

We propose an optical depth encoding scheme for detecting image tampering, embedding imperceptible depth signatures during capture.

An AI-based system for automatic, objective assessment of motor disease severity in Parkinson's disease patients using video analysis.

We propose FlatTrack, a lensless gaze tracking system using ultra-thin cameras, enabling near-eye gaze estimation with a large-scale dataset.

We propose a radiomics-based approach for classifying coronary artery calcification risk from non-gated CT scans without requiring ECG gating.

We propose a double-mask lensless camera that performs encryption in the optical domain.

We propose a novel coded-aperture based dual pixel camera with superior depth and RGB imaging capability.

We propose a learning-based framework for designing phase-masks for ultra-thin lensless cameras.

We propose a novel physics-inspired neural network for fast and accurate RGB-D estimation from single-shot lensless capture.

We propose an unrolled physics-based network for Fourier Ptychographic reconstructions that outperforms existing methods for low dynamic range sensors.

We propose a general learning based framework to recover photorealistic scenes from lensless captures for both separable and non-separable forward models.

We propose a learning based framework for designing the optical and analog components of a privacy enhancing camera.

We propose a learning based solution to recover photorealistic scenes from globally multiplexed lensless measurements.

Website credits to Jon Barron