"Alerted.org

Low-field MRI has been gaining interest due to its potential to significantly reduce the cost of diagnostic imaging. However, images produced by low-field scanners often suffer from noise and reduced quality compared to those from high-field MRI devices. A promising direction in current research is the use of generative AI to denoise and enhance low-field MRI scans, mapping them to their high-field counterparts. A central challenge in this line of work is the scarcity of paired low and high-field MRI datasets. Low-field scanners are relatively uncommon, which limits both the overall availability of low-field scans and, more critically, the amount of paired data required for supervised learning. This data bottleneck slows progress in downstream applications such as:

1. Converting low-field MRI into diagnostically comparable high-field MRI interpretation by radiologists.

2. Performing accurate predictions on under-sampled low-field MRI using advanced machine learning techniques.

This research project aims to develop accurate methods for synthesizing realistic low-field MRI scans from existing high-field MRI data, as if they were truly acquired on low-field devices. This would alleviate the paired-data bottleneck and unlock progress for multiple downstream tasks. Several complementary approaches will be explored:

i. Inverse problem formulation: Recasting high-to-low field MRI conversion as an ill-posed reconstruction problem.

ii. Style transfer methods: Treating low-field characteristics as a domain-specific “style” applied to high-field scans.

iii. Noise modeling: Leveraging the fact that MRI data is subject to Rician-distributed noise during k-space sampling, and explicitly learning generative models of this noise.

iv. Physics-based simulation: Employing analytical methods and MRI physics simulators to generate synthetic low-field data grounded in scanner physics.

Approaches (i)–(iii) will primarily leverage generative AI models such as diffusion models and normalizing flows, while (iv) will build on domain knowledge from MRI physics. A key challenge is the scarcity of paired training data. However, MRI data is highly structured, which can be exploited. The successful completion of this project will likely combine generative modeling techniques with domain-inspired scientific priors, enabling the creation of realistic low-field data and improved methods for MRI reconstruction and diagnosis.

Responsibilities

• Implement, train, and evaluate a variety of discriminative and generative deep learning models.

• Conduct large-scale experiments on GPUs/HPC clusters.

• Literature survey to understand the problem and the prior work.

About the Role

In this role, you will be participating in research projects at the intersection of Machine Learning, and Healthcare. Dr. Sumit Chopra, Associate Professor of Computer Science and Data Science at NYU Courant Institute of Mathematical Sciences and Director of Machine Learning in the Department of Radiology at NYU Grossman School of Medicine, will supervise your work. In compliance with NYC’s Pay Transparency Act, the hourly rate for this position is $26.00 per hour.

Bachelor's degree (Master's preferred)

Required Skills

• Python programming language

• Machine learning libraries such as PyTorch, Numpy, etc.

• In-depth knowledge of standard signal processing concepts

• (Preferred) MRI and the physics behind it

Upload your CV and a cover letter in Interfolio

For people in the EU, click here for information on your privacy rights under GDPR: www.nyu.edu/it/gdpr

NYU is an Equal Opportunity Employer and is committed to a policy of equal treatment and opportunity in every aspect of its recruitment and hiring process without regard to age, alienage, caregiver status, childbirth, citizenship status, color, creed, disability, domestic violence victim status, ethnicity, familial status, gender and/or gender identity or expression, marital status, military status, national origin, parental status, partnership status, predisposing genetic characteristics, pregnancy, race, religion, reproductive health decision making, sex, sexual orientation, unemployment status, veteran status, or any other legally protected basis. All interested persons are encouraged to apply for vacant positions at all levels.

Sustainability Statement

NYU aims to be among the greenest urban campuses in the country and carbon neutral by 2040. Learn more at nyu.edu/sustainability