Accelerating the diagnosis of bowel disease using AI

Millions of people every year suffer from bowel disease; and while treatments are advancing, the process to diagnose these kinds of diseases can be very labor-intensive for specialist doctors. Now, using artificial intelligence (AI), we can dramatically accelerate processes, free-up precious clinician time, and aid faster accurate diagnosis.

Given the inaccessibility of the bowel and the nature of these conditions, the output from an endoscopic pill – the commonly-used procedure for visualizing the small intestine – is around eight hours of video that must be watched and interpreted by a medical expert.

While senior clinicians can use all their experience to scan the video to find what they need, having to spend even an hour on this relatively mundane task is too much; and, inevitably, this process is prone to a margin of human error. It’s easy to imagine that for a handful of patients, the equivalent of a whole day of highly valuable clinician time could be spent each week just scanning through videos. This is where AI can make a major contribution, by focusing a clinician only on the frames of the video that they need to examine.

Improving patient, clinical and financial outcomes

This was the hypothesis that Marco Ragogna, an Atos AI Senior Expert at our Life Sciences Center of Excellence based in Italy, has been testing – and the results of a Proof of Concept are impressive. By training an AI system only to highlight the relevant frames, it is possible to reduce the time required so that the clinician only needs to spend around five minutes out of every hour previously spent looking at videos to detect any possible condition.

The benefits of this are significant, improving patient outcomes with faster diagnoses and fewer human-related mistakes, and improving clinical and financial outcomes with far more procedures reviewed efficiently and accurately in the same timeframe.

Requirements for implementation

For implementation, gaining access to the necessary repositories of videos is straightforward given the wealth of output produced in hospitals already. These videos need to be well curated and annotated with the help of medical staff; the relevant frames then need to be cleansed and pre-processed before being used to build, train, test, validate and optimize an AI model, with the system emulating how the human brain works to analyze images. In terms of the technology that’s required, videos require more pre-processing than still images, with the AI modelling requiring the huge computing power that’s available using high performance computing (HPC), either on premise or in the cloud. Inference from the images requires a fast local response, which means installing an edge computing server located in the clinical setting.

Artificial intelligence improving patient outcomes with faster diagnoses and fewer human-related mistakes

Proven solution and results

This combination of AI, HPC and edge computing has been proven through the Proof of Concept, which was developed at Atos’ HPC, AI and Quantum Life Sciences Center of Excellence in Cambridge. This successfully optimizes endoscopic capsule analysis and the associated interpretation process in a way that accelerates the whole procedure by a factor of 12 consistently and accurately. One of the founding principles of the Center is that, together with research and innovation, close collaboration with our customers and others in technology, research and clinical practice is vital. Given the results of our Proof of Concept, a number of hospitals are engaging with us to explore how to advance this solution in their settings.

By Natalia Jiménez Lozano, Director, Atos HPC, AI & Quantum Life Sciences CoE

Posted on February 25