Research Projects

Addressing scientific challenges often involves extracting and interpreting information from domain-specific signals, typically framed as inverse problems. The explosion of data has intensified this need, driving progress in artificial intelligence (AI) and computational imaging. By integrating cutting-edge algorithms with novel imaging systems, computational imaging extracts actionable information beyond the capabilities of traditional methods. This project builds on these principles to develop advanced signal processing and AI techniques for enhanced computational imaging, with a focus on imaging sensors and biomedical applications, aiming for broad interdisciplinary and societal impact. A key objective is imaging sensor optimization, which integrates innovations in optics, sensor design, and image processing, to enable real-time, high-quality imaging. This is crucial for applications ranging from medical diagnostics to autonomous vehicles. While AI has enhanced sensor performance, challenges such as high latency persist. This project will address these through novel sensor encoding strategies and fast decoding algorithms. Specifically, this project will develop advanced deep learning models, focusing on diffractive deep neural networks (D2NNs). D2NNs offer ultrafast, light-based data processing without electronic computation. The ultimate goal is to create a fully integrated, real-time imaging system with optimized sensor performance. In biomedical imaging, computational techniques are vital for improving image reconstruction and diagnostic accuracy. Early detection of fatty liver disease—impacting nearly 30% of the global population—is hindered by invasive and expensive diagnostic methods. Quantitative ultrasound (QUS) biomarkers offer a promising, accessible alternative. While deep learning has shown potential in addressing limitations of traditional ultrasound models, its adoption remains limited due to computational demands. By integrating traditional signal processing with advanced AI, this project aims to enable real-time, robust, and accurate fatty liver diagnosis. A key focus will be clinical translation, supported by uncertainty quantification using methods such as Bayesian neural networks. Innovations in sensor design with low latency are expected to significantly advance both camera sensor technology and biomedical imaging. As for biomedical imaging, the developments will facilitate the determination of optimal quantitative ultrasound (QUS) thresholds tailored to the Indian population, enabling integration into commercial scanners. Incorporating uncertainty measures will enhance clinician confidence and support robust clinical adoption. In conclusion, this interdisciplinary research will push the boundaries of imaging technologies, offering innovative solutions to longstanding challenges and opening new avenues for exploration and application.