About Me

I completed my Ph.D under Dr. Phaneendra K. Yalavarthy in field of Medical Imaging at Department of Computational and Data Sciences in Indian Institute of Science, Bangalore. Link to my RESUME is HERE.
The website of Medical Imaging Group of Indian Institute of Science can be found HERE I did my Bachelor's in Technology from Delhi Technological University in 2016. After that, I worked as Senior Engineer in Bajaj Automobile Ltd. for close to two years. Currently, I am working as a Postdoctoral Researcher and technical subgroup leader Prof. Dr. med. Philipp Vollmuth's group (Division of Computational Radiology and Clinical AI) in Universitätsklinikum Bonn, Germany.

I am the recipient of the prestigious Prime Minister’s Research Fellowship in August 2020

Projects

VTDCE-Net

A Time Invariant Deep Neural Network for Direct Estimation of Pharmacokinetic Parameters from Undersampled DCE MRI Data

Description

The objective of this work it to propose a robust time and space invariant deep learning algorithm to directly estimate pharmacokinetic/tracer kinetic (PK/TK) parameter from undersampled dynamic contrast-enhanced (DCE) MR images. DCE MRI consists of 4D (3D-spatial + temporal) data and is used to estimated 3D (spatial) tracer kinetic maps. Existing deep learning architectures for this task are either not invariant to temporal dimension or to both temporal and spatial dimensions. We proposed a deep learning algorithm that is invariant to training and testing in both temporal and spatial dimensions, i.e we can train the network on dataset consisting 'x' time samples and test in on dataset consisting 'y' time samples. Our experiments found that VTDCE-Net performs better than the Total Variation scheme on both breast and brain datasets for estimating capillary permeability and blood vessel volume fraction for all undersampling rates. This work is published in Medical Physics Journal [3]

SpiNet

Schatten p-norm Regularized Medical Image Reconstruction [Github Page][Results]

Description

The objective is to give a deep learning architecture for solving inverse problems which explicitely incorporates the forward model and uses noise statistics as prior. The noise prior is learnt using Deep Learning and the novelty of this architecture is that it can enforce any $p$ norm where $0 < p \leq 2$ and $p$ can be kept fixed or estimated from the data (a trainable parameter). This work is published in Medical Physics Journal [2]

DCE MRI

Comparison of iterative parametric and indirect deep learning-based reconstruction methods in highly undersampled DCE MR Imaging of breast [Github Page][Results]

Description

The purpose of the project was to compare the performance of iterative direct and indirect parametric reconstruction methods with indirect deep learning based reconstruction methods in estimating tracer-kinetic parameters from highly undersampled DCE-MR Imaging breast data and provide a systematic comparison of the same. This work is published in Medical Physics Journal [1]

Image Denoising using CNN

E9 253:Neural Networks & Learning System Course Project [Github Page] [Results]

This project is done along with Ms. Ocima Kamboj [Website]

This project was part of course E9 253: Neural Networks and Learning Systems the goal was to implement Deep Learning Denoiser using residual learning. We attempted to replicate the results of Zhang et.al. in their paper Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising and experimented with L2, L1 and SSIM loss functions and their combinations. We trained the network for gaussian denoising, deblurring, JPEG deblocking and image super resolution

Assignments

DS200:Research Methods

Module 4 [Github Page] [Results]

Description

The objective is to use data visualization tools available in the python library matplotlib. The data used is the state-wise Police Complaint Registration Statistics of 2009 of all Indian States and U.Ts. It is available at https://www.data.gov.in and was published by http://www.ncrb.gov.in, both part of Ministry of Home Affairs of Indian Government. The data gives the statistics of the method of complaint registration (oral/written/helpline/suomoto by police) and the category in which complaint is registered (IPC or SLL).I used scatter plot, box plot and bar chart to analyse the data and draw inference. The links to the Result and Github repository are given above.

Teaching Duties

I have assisted in teaching of the following courses:

Course	Instructor	Institute	Year
DS 288: Numerical Methods	Prof Phaneendra K Yalavarthy	Indian Institute of Science	Aug 2019 - Dec 2019
Biology for XI standard	Mrs Priyanka Sharma	Kendriya Vidyalaya, IISc	Jan 2021 - June 2021
DS 288: Numerical Methods	Prof Phaneendra K Yalavarthy	Indian Institute of Science	Aug 2021 - Dec 2021

Publications

[1]. Aditya Rastogi and Phaneendra K. Yalavarthy, ``Comparison of iterative parametric and indirect deep learning-based reconstruction methods in highly undersampled DCE‐MR Imaging of the breast,” Medical Physics (2020) [This work is the first comprehensive comparison of compressive sensing reconstruction methods with model based deep learning methods for the breast perfusion imaging and shows that deep learning methods are sub-optimal at higher undersampling rates.]

[2]. Aditya Rastogi and Phaneendra K. Yalavarthy, ``SpiNet: A Deep Neural Network for Schatten p-normRegularized Medical Image Reconstruction,” Medical Physics (2020), [This work is first-of-its kind in proposing ageneric Schatten p-norm (0< p≤2) regularization based deep learning network for medical imagereconstruction, wherepis a trainable parameter (chosen automatically).]

[3]. Aditya Rastogi Arindam Dutta, and Phaneendra Kumar Yalavarthy, ``VTDCE‐Net: A time invariant deep neural network for direct estimation of pharmacokinetic parameters from undersampled DCE MRI data.” Medical Physics (2022).

Miscellaneous

Comics

Favourite comic strips

Favourite quotes from TV Series

Yes Minister, S1E1

Jim Hacker: I’d like a new chair. I hate swivel chairs.
Bernard Woolley: It used to be said there were two kinds of chairs to go with two kinds of Minister: one sort folds up instantly; the other sort goes round and round in circles

Yes Minister, S2E2

[Bernard explains to the Minister the honours available to senior Civil Servants.]
Hacker: Well, what has Sir Arnold to fear, anyway? He’s got all the honours he could want, surely?
Bernard: Well, naturally he has his G.
Hacker: G?
Bernard: Yes; you get your G after your K.
Hacker: You speak in riddles, Bernard.
Bernard: Well, take the Foreign Office. First you get the CMG, then the KCMG, then the GCMG; the Commander of the Order of St Michael and St George, Knight Commander of St Michael and St George, Knight Grand Cross of St Michael and St George. Of course, in the Service, CMG stands for “Call Me God,” and KCMG for “Kindly Call Me God.”
Hacker: [chuckles] What does GCMG stand for? Bernard: “God Calls Me God.”

Thin Blue Line, S2E2

Inspector Fowler: Appearances, as we shall see, are like bus timetables: often highly misleading.