New Research to Improve AI Diagnosis from Chest X-Rays
In a world-first study, a team from the Australian e-Health Research Centre (AEHRC) at the national science agency the Commonwealth Scientific and Industrial Research Organisation (CSIRO) compared the accuracy of different AI models in interpreting chest X-rays, the Xinhua reported.
They found that by using the optimal combination of an encoder and decoder, automated AI diagnoses of heart and lung conditions from X-ray images can be made 26.9 percent more accurate.
Current AI X-ray report generation technology uses an encoder to interpret images and a decoder to produce a report. The AEHRC research was the first globally into which encoder and decoder are best for accurate diagnosis, the CSIRO said.
"AI has the potential to improve health services, and in particular better support health professionals by easing their burden and workload of current non-automated practices," Aaron Nicolson, a CSIRO Research Scientist and lead author of the study, said in a media release.
"Automated report generation for X-rays could reduce clinician burnout and create space for them to provide more robust patient care. The research demonstrates the future potential to better support clinicians."
In addition to testing different encoders and decoders, the AEHRC team also employed a method known as "warm starting," whereby the knowledge learned by an AI model from doing one task is applied to improve a second task.
They found that the model was able to identify some lung abnormalities, such as pleural effusion -- a build-up of fluids -- more consistently than others, such as lung lesions.
The team will next seek to improve the model so it can accurately identify most conditions consistently before the technology is used in clinical settings.
4155/v
