Artificial intelligence used to predict patient death

By / 1st of June, 2017

ARTIFICIAL intelligence is being used to predict death with surprisingly accurate results.

The computer program designed by researchers at the University of Adelaide in South Australia, studied images of organs in 48 patients and was able to predict which patients would die within five years with almost 70 per cent accuracy.

The technology uses deep learning to recognise complex imaging appearances on CT scans.

The study, which also included researchers from Portugal, was published today in the Nature journal Scientific Reports.

Lead author Luke Oakden-Rayner said the AI had the potential for early diagnoses of medical conditions such as heart disease.

“There is only a certain amount of information humans can see when looking at scans but research suggests there is more we can see using computers,” he said.

“Our program allows us to look at medical outcomes rather than diagnoses, which is able to tell us how patients are going to do in the future instead of how they are doing right now.

“Using deep learning we are able to understand and analyse organs in a way doctors can’t do at the moment.”

img - industries_technology_health_AI predicts death_bImages at the level of the proximal left anterior descending coronary artery, with the most strongly predicted mortality and survival cases selected by averaging the predictions from the deep learning and engineered feature models. The mortality cases (left side) demonstrate prominent visual changes of emphysema, cardiomegaly, vascular disease and osteopaenia. The survival cases (right side) appear visually less diseased and frail.

Dr Oakden-Rayner said there had been no previous research, which relied on image data alone to calculate longevity.

The program is fed various CT scans and calculates the brightness of pixels on each photograph.

The pixels are all given numerical values to represent the brightness and the system analyses a pattern across those numbers without knowing precisely what the image is.

Important factors such as age, gender and genomics were excluded from the research to ensure the results were solely based on image analysis.

According to Dr Oakden-Rayner, contemporary studies that included age and gender as indicators of mortality had accuracy levels of about 60-70 per cent.

He said this implied that if the new AI took into account those extra data fields, it would significantly increase its success at predicting death.

While the researchers could not identify exactly what the computer system was seeing in the images to make its predictions, the most confident predictions were made for patients with severe chronic diseases such as emphysema and congestive heart failure.

“We already have thousands of cases now and are working through them with some preliminary results showing the things we would expect to see,” Dr Oakden-Rayner said.

“This has strong implications for being able to tailor treatments to the patient and could also be applied to almost any type of medical outcome … not only diseases but things like surgery success or how likely someone is to fall over and break their hip.”

Dr Oakden-Rayner worked closely with leading deep learning researcher in the field of medical science Associate Professor Gustavo Carneiro from the University of Adelaide.

Together they wrote the program and hope to apply the same techniques to predict other conditions such as the onset of heart attacks.

South Australia’s capital Adelaide has three long-standing public universities, Flinders UniversityUniversity of South Australia and the University of Adelaide, each of which are consistently rated highly in the international higher education rankings.

Key contacts

Luke Oakden-Rayner School of Public Health Radiologist
61 434 166 627 luke.oakden-rayner@adelaide.edu.au