COMP0016 Team 15

There is a growing demand in the NHS for using machine learning to aid in diagnosis. Classifiers are being built by data scientists, but they need clinicians to test them in various ways at each stage from initial prototypes to final deployable tools. Currently, there is no easy way for different machine learning classifier models to be accessed and trained for testing by clinicians working with data scientists in multi-disciplinary groups. Current solutions host datasets, but what is provided by data scientists are either research papers, or at best source code. Training models or requesting code takes time and recourses, and you can run into problems with dependencies, versioning, and lack of expertise. This is the problem we set out to solve.

Our solution will provide a closed and secure system for hosting of datasets and trained machine learning models which when uploaded are dockerised with their given dependencies. These docker images can then be downloaded, run, and tested with a provided frontend. Feedback can then be provided on these models to aid in collaboration and development.

We have completed all the requirements set out by our client. Our system is deployed, and functioning on a Linode server at: http://176.58.116.160/, ready to be tested by clinicians and data scientists. Our source code is hosted on GitHub and shared with our project partners with full documentation and a deployment manual for running locally or hosting on Linode. We have a set of possibilities for future work in our evaluation section and have focused on making our code as reabable and extendable as possible, with the hope that this can be built on by future systems engineering teams or even developers in the NHS.