Computer-aided diagnosis and its challenges

Prashanna Khwaounjoo has an interesting life.

He’s helping to prototype data analytics from medical devices, needs unhealthy people to work with, and is working on the trust element with computer-aided diagnosis.

Prash graduated in biomedical-engineering from the Auckland Bioengineering Institute, University of Auckland where he came him into contact with the MedTech Centre of Research Excellence.

“With our collaborators at the ABI, MicroVision, a displacement algorithm that could measure pressure wave forms around the neck was developed. This application opens a multitude of prospects. A contact in Turkey has a high-end ultrasound machine that could, for example, produce 3D analysis of veins in the neck for cholesterol plaque. There’s an opportunity to correlate these images with the app for better diagnosis.”

Now at the University of Otago, he’s focusing on translational and data analytic projects and also mobile health apps. While a large part of his work centres on neuroscience and modulation of the vagus nerve in Parkinson’s patients, his key interest is devices.

“I’m looking at how outputs from medical devices help with diagnosis, and part of this involves helping to prototype data analytics where we take those outputs, assess the parameters and trends, and use machine learning to improve patient outcomes,” says Prash.

But there are challenges. “It’s easy to get healthy people, but you also need to test these ideas on unhealthy people, and then you have to gain access to the clinicians.” He also has to work out how to balance the usefulness of an app against the expertise of the clinician so there is demonstrable benefit.

“There’s a big question around how much you can trust data from an app, and we must ensure clinicians see apps as supplementing their work rather than overriding it. We want them to see it as something that can increase their confidence.”

Prash cites mole mapping as a potential good collaboration between machine and clinician. “While at MoleMap we were developing deep learning and AI to classify lesions, to increase the confidence around whether it’s benign or cancerous. We were creating an algorithm that in future will increase a dermatologist’s accuracy, speed up the classification process and increase confidence in the diagnosis.”

Again, there are challenges, such an as creating an app with high enough accuracy that it does speed up the process, leaving dermatologists and their staff to focus on where they’re most needed.

He doesn’t believe we or should rely entirely on computer diagnosis. A computer may provide a quantitative diagnosis, but a clinician adds the qualitative. Prash sees the potential for computers and apps as very much based around them being used as assistive technology for the clinician.

By Prue Scott

Prashanna Khawounjoo