Rupesh Gupta Email & Phone Number

Ann Arbor, Michigan, United States

I work on developing machine learning and image processing solutions that enable point of care CT imaging for various applications.

Ann Arbor, Michigan, United States

Guide : Prof Jeffrey Fessler, Department of Electrical and Computer EngineeringProf Yuni Dewaraja, Department of RadiologyI'm working on implementing deep learning methods for low-count quantitative image reconstruction in PET (Positron Emission Tomography) and SPECT (Single Photon Emission Computed Tomography) imaging.Full count PET/SPECT scans in clinical systems take about 30 minutes. I've been working on improving the quality of reconstruction for low-count imaging, which can be used to reduce scan-time or image patients in pre-therapy. I've designed and implemented an iterative reconstruction method, called BCD-Net that solves the imaging problem using a trained Convolutional Neural Network regularizer.This network, albeit simple enough to avoid learning the imaging physics, has shown to improve the reconstruction accuracy by 42%, measured by activity recovery while also ensuring confidence in recovery through modelling of image physics.This model belongs to the broader category of Plug-n-play methods for iterative reconstruction.Currently I am working on developing a policy network for automatic search of parameters, which can be implemented using deep Reinforcement learning models.

Bangalore

As a research intern, I developed an unfolded-FISTA deep neural network for parameter estimation in Finite Rate of Innovation (FRI) signals. This model is useful in improving the depth resolution in Ultrasound imaging. Backtested parameters for Ultrasound signal had around 5% lower minimum mean square error at all noise levels compared to benchmark and baseline analytical models.

Tiruchchirappalli Area, India

Under the guidance of Prof. P. Palanisamy, I worked on bachelor thesis project on compression of electro-cardiogram (ECG) signals using compressing sensing model, useful in Tele-cardiology applications. I implemented a modified adaptive dictionary learning based reconstruction framework using MATLAB. The model effectively utilizes mutual coherence reduction techniques to enhance the signal sparsity in reconstruction. Testing on MIT-BIH Arrhythmia database indicated 18% reduction in percentage-root mean square-difference compared to baseline models. The work has been accepted in an IEEE conference (ICCSP'20).

Berlin Area, Germany

Implemented a neural network on a Calliope-mini micro-controller to estimate the height of a person using accelerometer data collected from hand gesture movements. Also, demonstrated optimization in data transmission at wireless sensor nodes.

Bengaluru Area, India

Developed a MATLAB code to analyze the depth resolution of FRI reconstruction method in estimating the layer thickness of a two-layer specimen using Ultrasound pulse. Analyzed and explained the resolution capability using 4th order NP-FRI model for real US data collected at various noise levels.

Mumbai, Maharashtra, India

Worked on Single Board Heater System (SBHS) Remote Triggered Virtual laboratory. SBHS virtual lab is an initiative to help students, teachers and in general enthusiasts to perform experiments in Instrumentation and Control labs virtually. We proposed a new load-balanced architecture for this system, involving slave raspberry pi controllers and single master server, which improves the scalability and robustness of the setup.

Master'S Degree, Ee: Signal And Image Processing And Machine Learning, Gpa: 4.0/4.0

My focus is on design of theoretical well-founded algorithms using DNNs for image processing applications with emphasis on medical imaging.

Bachelor’S Degree (Honors), Electronics And Communications Engineering, Gpa: 9.94/10.0, Gold Medalist

Activities and Societies: Manager at Pragyan Workshops team, Pragyan NIT Trichy

High School, High School/Secondary Diplomas And Certificates, 4Th-12Th

Activities and Societies: School Captain in 9th grade; active member of the Science club