Abstract
Purpose:
Detecting the presence of cataracts can help restore a patient’s vision but also hint at a patient’s diabetic state or other underlying pathologies. Given the huge backlog of cataracts worldwide, rapid determination of the patients in most need for surgery is an important public health task. Consequently, we present a quick and simple method to determine the presence of a cataract based on automated imaging of the retina.
Methods:
The method is based on the comparison of a photo of the fundus obtained using a standardized setting of a digital non-mydriatic fundus camera, to that of a similar photo of a non-cataractous eye. The inexperienced observer who has no previous knowledge of cataract pathology or grading systems completes a brief training session. Right to left, preoperative and postoperative eye and eyes from different patients comparisons are included in the training database as these discrimination tasks are all relevant to the goal of referral and allocating resources. Following the training session the observer is tested with a different set of images to determine the effectiveness and efficiency of this method.
Results:
The recognition rate of minimally trained technicians was generally quite high (above 95%) with all observers classifying 100 hundred pairs within less than 5 minutes.
Conclusions:
Our methodology demonstrates that cataracts can be identified by minimally trained technicians based on the automated acquisition of a fundus photo. This has implications for public health efforts and telemedicine programs that can contribute to eradicate a reversible cause of blindness worldwide. One novel aspect in our methodology is the minimal need for training and the provision of a computerized learning environment where rapid recognition of cataractous states can be learned by an untrained remote observer in less than half a day.
Keywords: 445 cataract •
550 imaging/image analysis: clinical