Aussie AI takes on radiology diagnostics


James Riley
Editorial Director

An Australian healthcare joint-venture between artificial intelligence startup Harrison.ai and local radiology company I–MED Radiology Network has launched a chest X-ray diagnostic tool on the same day a peer-reviewed diagnostic accuracy study of the product was published by The Lancet Digital Health.

The joint-venture – Annalise.ai – unveiled its chest X-ray AI solution Annalise CXR, which detects 124 clinical findings and is a decision support tool for radiologists and clinicians. The next most comprehensive CXR AI product detects just 75 clinical findings, with most CXR AI products limited to fewer than 15 findings.

The study published in The Lancet Digital Health found that when used as an assist device, Annalise CXR significantly improved the ability for radiologists to perceive 102 chest X-ray (CXR) findings in a non-clinical environment, was statistically non-inferior for 19 findings, and no findings showed a decrease in accuracy.

Aengus Tran
Co-founder Aengus Tran: AI diagnostics for radiologists

The study also assessed the standalone performance of the model in a non-clinical environment against radiologists in identifying chest x-ray pathology, as well as investigating the effect of model output on radiologist performance when used as an assist device.

Annalise CXR’s AI model classification alone was significantly more accurate than unassisted radiologists for 117 (94 per cent) of 124 clinical findings predicted by the model and was non-inferior to unassisted radiologists for all other clinical findings.

The performance evaluation points to a breakthrough product. “The ability of the AI model to identify findings on chest x-rays is very encouraging,” says Dr Catherine Jones, a thoracic radiologist and Chest Lead at annalise.ai.

“Radiologists and non-radiology clinicians incorporate clinical factors into decision making, but ultimately rely on perception of findings to underpin our clinical interpretation,” she said.

“Developing and validating a comprehensive AI model for chest x-rays required a careful, detailed approach based on robust methodology and focus on quality labelling, training, software development and thorough evaluation prior to clinical deployment.”

The Annalise CXR AI model was trained on 821,681 chest X-ray images from 520,014 studies across 284,649 patients. The study assessed the performance of the radiologists alone and the same radiologists when using the AI model as an assist device when identifying pathology in a chosen dataset.

Twenty radiologists each reviewed 2,568 CXR studies both with and without the assistance of the Annalise CXR model, allowing adequate time between both arms of the study to minimise bias. Gold-standard ground truth labels were obtained from the consensus of three sub-specialty thoracic radiologists with access to reports and clinical history.

Annalise.ai co-founder and chief executive Dimitry Tran said Annalise CXR would provide significant benefits to patients and to healthcare professionals.

“A major challenge facing global health systems is that the number of scans requiring clinical interpretation is growing at a much greater pace than increases in the number of radiologists to interpret them,” Mr Tran said.

“Annalise CXR seamlessly integrates with regular workflows, highlighting findings on chest X-rays for review by the radiologist.

“We hope that the solution will increase radiology capacity, thereby reducing turnaround time [and] improving interpretation quality by providing clinicians with another set of eyes, and reducing the risk of backlogs,” he said.

Annalise.ai is located in Australia, Singapore and the UK.

Do you know more? Contact James Riley via Email.

Leave a Comment

Your email address will not be published.

Related stories