This has no doubt been the decade of AI, in terms of both the influx of AI-powered products as well as AI systems, and though opinions are divided about the extent of its applicability, in the area of medical diagnosis support has been growing steadily. For this reason, recent news of developments from the deep learning Google subsidiary DeepMind about diagnosing eye diseases should come as no surprise.
DeepMind partnered with University College London (UCL) and London-based Moorfields Eye Hospital, on the joint project which involves the development—and later clinical trials—of new methods aimed at swift and accurate diagnosis of some of the diseases which threaten our sight the most.
With a process that introduces “near-infrared light off of the interior surfaces of the eye”, data from 3D retinal scans is created in roughly 10 minutes.
The extensive trials involved upwards of 7,500 patients as well as nearly double that number of OCT scans.
Through careful analysis which included learning the anatomy of the eye and using this information to offer informed opinions, the system learned diagnosis procedures.
Details about the promising research were shared this week in an article, titled “Clinically applicable deep learning for diagnosis and referral in retinal disease”, in Nature Medicine.
The innovations offer a comprehensive method for eye condition diagnosis, which mainly includes optical coherence tomography (OCT), a method for generating 3D images, which in itself is accurate. However, in terms of the time and skill set required to effectively analyze and interpret the data, the team sees its shortcomings.
DeepMind co-founder Mustafa Suleyman said of the transformative potential of the technology in medicine:
“Our research with DeepMind has the potential to revolutionize the way professionals carry out eye tests and could lead to earlier detection and treatment of common eye diseases such as age-related macular degeneration. With sight loss predicted to double by the year 2050[,] it is vital we explore the use f cutting-edge technology to prevent eye disease.”
Not shying away from controversy, DeepMind also made reference to the ‘black box’ (in terms of how it impacts AI, the term ‘black cloud’ at times seems more appropriate) issue that leads to skepticism and criticism against AI systems.
Their approach involves the use of two complimentary neural networks, —one for comprehensive map analysis, and the other for generating a map of the OCT scan, DeepMind has termed the classification method and segmentation network.
Their hope, then, is to both create a stronger and more effective means of diagnosis, but in addition, provide more transparency about how conclusions are reached, which is one of the biggest stumbling blocks for AI system acceptance.
With each AI systems innovation, and along with overwhelming support and recognition of its benefits, the resistance to its use in the realm of medical diagnosis will fade away and it will gain more and more acceptance as a trusted standard.
The research was published in Nature Medicine.