Web Interface and Cloud Deployment for MRSNet

Uday Motiram Sawant


Supervised by Frank C Langbein; Moderated by Dr Soumya Barathi

MRSNet is a deep learning tool to quantify metabolites in magnetic resonance spectroscopy (see https://qyber.black/mrs/code-mrsnet - more code, by now also ported to python3 with many extensions and clean-ups, will be available if you choose to join the project; we are about to release V2). This is relevant for medical diagnosis as well as understanding of biochemical processes. The aim of this project is to develop a UI to make MRSNet more usable in the relevant application domains and simplify its deployment as a cloud service. At the moment it is a python library with a command line frontend, aimed at exploring the machine learning approach rather than being used for the application. For the UI, keep in mind that potentially many spectra need to be quantified, many machine learning models may be used and the accuracy of the quantification will have to be verified. We envisage a web user interface with backend GPU servers to run the analysis and potentially also training. Backend analysis services would ideally be configurable by the user to choose their resources, hosted via containers (as we do not have the resources to offer this as a general service on the web). You can use any suitable programming language and framework, but the code must be released under the AGPL v3 or later. GPU resources are required (for prediction, without training, this can be a small GPU) and resources for this are available in the Linux lab or from SCW; potentially we can explore hosting this on AWS, but we need to check if resources for this are available from an SCW trial project.

Final Report (11/09/2023) [Zip Archive]

Publication Form