COVID-19 collab: A platform for COVID-19 machine learning based diagnostic tools

Roger Noble

To assist with the current pandemic, we are facilitating a community led effort to develop a highly accurate diagnostic tool, based on machine learning techniques, to aid in the detection of COVID-19 from patient X-Rays and other imaging modalities. The proposed platform will be based on our cloud platform (Zegami) in addition to a prototype machine learning model we've already developed for the detection of COVID-19 within x-rays. This builds upon a prototype diagnostic tool that Zegami has already developed, which demonstrates that AI can be used to differentiate between COVID-19, pneumonia and healthy patients with an accuracy rate of 70%.
Our prototype model was trained on several freely available datasets that have been collected from all over the world. The training data was rapidly made available by academics and clinicians by both the NHSx and others as a response to the pandemic. One of the problems that we’ve found when working with these data sets, was the large variability between each of the images. These were differences between techniques of individual operators, procedures and equipment, which can cause a machine learning model to make incorrect assumptions. Zegami played a pivotal role in cleansing the data, in addition to visually validating the model once it's trained. In summary, this allowed us to rapidly iterate on the model by having a visual feedback loop into how the model was performing, saving considerable weeks of effort.
Working with the NHSx and their recently released COVID-19 dataset, and in partnership with the National Consortium of Intelligent Medical Imaging (NCIMI), Zegami will develop an open platform for researchers to develop and validate their machine learning models. The platform will be freely available for anyone working on COVID-19 related projects.
This project wouldn’t be successful without the willingness for researchers and universities to work with industry. We’ve experienced many renewed conversations over the course of the last month that had previously grown stale due to apprehension with working with industry.