From accurate clinical diagnosis to autonomously operating surgical robots, AI has more to offer than we have asked for.
The ophthalmic practice had changed with the advent of every new technological breakthrough. The intra-ocular lens in sixties and minimally invasive cataract surgeries in nineties had a significant impact on the practice. Both of these technologies revolutionized the clinical out come in the terms of speed and accuracy of visual restoration making it the ophthalmic surgeries, fast and predictable.
The idea of artificial intelligence (AI) to correctly identify diseases and then effectively treat patients as good as doctors do might be possible in the future. That idea will definitely take some time to take a shape, till than an ophthalmologist will be an indispensable cog in the wheel of patient care. Not many of us will reject the possibility of this revolution but we do believe that this is not happening any time soon.
Clinical diagnosis and management decisions
Reports about the upcoming AI related platforms in eye care, are painting an entirely different picture. Erping Long , China had tested An artificial intelligence platform for the management of congenital cataracts. They used a deep-learning algorithms to create AI agent involving different functional networks to perform three important task (i) identify potential congenital cataracts patients in populations (Screening) (ii) Comprehensive evaluations of disease severity (lens opacity) with respect to three different indices (opacity area, density and location) (Risk stratification) (iii) Provide the final treatment decision (Surgery or follow-up). The researcher explored its feasibility, versatility and utility in management of congenial cataract and compared its real-world performance with an individual ophthalmologist. The training data for this deep-learning network included 410 ocular images of CC of varying severity and 476 images of normal eyes from children, categorized by an expert panel.
The results of the study were simply amazing. The AI machine could distinguish cataract patients from healthy individuals with an accuracy of 98.87%. It could provide a treatment suggestion (surgery or follow-up) with an accuracy of 97.56%. They further validated its ability to identify the diseased patients in a realistic real-world ratio of rare-event disease to normal cases. In this case, the test data consisted of a total of 300 normal cases and 3 cataract cases of differing severity. The agent successfully excluded the normal cases, identified the three cataract cases and provided accurate evaluations and treatment decisions.
Its performance, when compared with individual ophthalmologists, was even more remarkable. In terms of accuracy, the AI agent out-performed individual ophthalmologists on every parameter analyzed, be it diagnosis or management decision. The ophthalmologists were categorized in three degrees of expertise (expert, competent and novice) for comparison. The AI agent performed better that experts in terms of false positives and missed detections.
From the AI context, the results were on expected line. As data heterogeneity is inevitable in clinical practice, and AI has inherent advantage over human in their ability to tackle multi source and wide-format data. Increasing computational speed, evolving deep learning algorithms and increasing sophistication of image recognition capabilities have tilted the balance simply in the favour of the AI machines only. In this particular case, the phenotype of congenital cataract has lot of variations , which makes the classification of congenital cataract images difficult, influencing decision-making and patient prognosis. The illumination intensity, angle and image resolution, eyelids, eyelashes and pupil size etc., make the identification peculiarly complex for an ophthalmologist. The AI machine with advantages of deep-learning algorithms, learns from the cumulative data and becomes more and more intelligent with increasing use and widening of the input base. Its predictions becomes ever more accurate with every possible cycle of use. Though validated in case of congenital cataract, similar collaborative platforms will soon be extended to the management of other diseases, and find validations in different clinical scenarios. Google DeepMind, in collaboration with the NHS, is working on a machine learning system which will eventually be able to recognize sight-threatening conditions such as wet age-related macular degeneration and diabetic retinopathy, from a digital scan of the eye. And very soon, we will have similar platforms for other diseases as well.
Available in market already available in market. Recently FDA has given clearance to an artificial intelligence-powered IDx-DR diabetic retinopathy detection system which is supposedly the first device cleared to diagnose a medical condition autonomously. What is more interesting is the fact that it is authorized to do so without requiring a review by a specially trained clinician. It will not only drastically increase the number of diabetic people receiving screenings but also reduce the requirement of an ophthalmologist. This has also received a green signal in Europe and the initial trails are been reported to have encouraging results.
The AI assisted robotic surgical devices setting newer standards in the field of ophthalmic surgery. R2D2, developed by Preceyes BV, a Dutch medical robotics firm can perform complex retinal surgeries like epiretinal peeling surgery. Though it was been guided by the surgeon during the surgery, the robot was capable of performing it autonomously as well. It is not the only robot which is capable of operating on human eyes. Axsis developed by Cambridge Consultants, can precisely navigate in a small space as eye and could also be used to operate senile cataract. What is more interesting is the facts that these robots will enable surgeons to perform surgical procedures which are not yet possible, like in sub retinal space and in and around blood vessels in the eye.
It is evident that the role of an ophthalmologist will be redefined in practice, by these upcoming technologies. Ophthalmologists will have to find their roles more important than diagnosis and surgery, as the AI will not have the element of human error and learning gaps. Now it is up to us to identify how we adopt ?