The OptimalAI team worked with IBM to alleviate the NHS's radiology problem by adapting the PowerAI platform to augment the skills of radiologists. NHS radiologists are now able to train algorithms on a dataset of millions of images with above 90% accuracy within half an hour.  This has dramatically reduced the number of scans and x-rays needed to be assessed by radiologists both in the UK and around the world.  Upon implementation, the platform immediately saved the NHS £116 million in outsourcing scans and staff overtime.

With approximately 70% of scans and x-rays typically classified as ‘normal’ and not requiring further investigation, PowerAI eliminates these from the radiologists’ workload. This leads to quicker feedback to clinical teams where no issues had been found and allows radiologists to focus on those scans that need further investigation.

PowerAI allows radiologists who've identified bottlenecks in radiology workflows to create powerful AI models by labelling images according to their specialist domain. Using a publicly available set of anonymized x-rays and scans radiologists can create a model that can rapidly classify whether an x-ray is ‘normal’ or needs further investigation. Obvious initial applications could be triaging of abnormal chest x-rays, breast mammograms or skeletal fracture detection.

AI models will always need human expertise to create the initial validation of images but Shah's work with IBM's PowerAI provides an environment that makes these easy to create without any prior programming knowledge, for the first time giving autonomy back to experienced clinicians who understand where AI can best augment workflow and are best served to use it. The unique platform provides the processing capacity based on IBM Power9 servers to enhance the time to accuracy.