A NEW computational data tool has been created to diagnose eye disease and pneumonia that is accurate and far quicker than traditional methods. The technology, developed by researchers from the University of California, California, USA, along with other colleagues, could be broadened to apply to other kinds of medical conditions in the future, including cancer.
Transfer Learning and Occlusion Testing
In the study, the team were able to analyse over 200,000 eye scans conducted with optical coherence tomography using an artificial intelligence (AI)-based convolutional neural network. Transfer learning, in which knowledge stored in a computer about how to solve one problem is applied to related problems, was used to enable the AI system to learn effectively with a much smaller dataset than in conventional techniques. Following this, the researchers added occlusion testing, whereby the computer shows specific areas in an image that were of the greatest significance in coming to its conclusions, increasing transparency and trust of the findings.
When used to diagnose two types of eye disease, the machine was found to perform to a similar level to a well-trained ophthalmologist and could take just 30 seconds to generate a decision on treatment referral, with an accuracy rate of over 95%. The technology was also utilised to diagnose childhood pneumonia by analysing chest X-rays; here it was found to have an accuracy of over 90% in distinguishing between viral and bacterial forms of the condition.
The speed and accuracy of this method means it has the potential to improve diagnosis and treatment in medicine, with effective treatment often being delayed by the time it takes to refer patients. Senior author Prof Kang Zhang, UC San Diego, commented: “AI has huge potential to revolutionise disease diagnosis and management by doing analyses and classification involving immense amounts of data that are difficult for human experts, and doing them rapidly.”
The authors believe this AI technology has the potential for plenty of other applications in medicine, including differentiating between benign and malignant lesions detected on scans.
The tool and data have been open-sourced so that other scientists can further improve, refine, and develop its potential.
James Coker, Reporter
For the source and further information about the study, click here.