Abstract
Purpose :
As trachoma is eradicated, skilled field graders become less adept at correctly identifying follicular trachoma (TF). Deciding if trachoma has been eradicated from a district or if treatment strategies need to be reinstated is of critical public health importance. Telemedicine solutions require connectivity, which can be poor in the resource-limited regions of the world in which trachoma occurs. Our purpose was to develop and deploy a deskilled, low-cost telemedicine solution to acquire and transmit high-quality tarsal plate images to a cloud-based virtual reading center using crowdsourcing for image interpretation.
Methods :
A hands-free digital imaging solution was developed to allow field workers to acquire high-quality images using both hands to evert and visualize the superior tarsal plate. A system was then developed to allow for in-browser local compression of images for transmission to a cloud-server to minimize the need for broadband internet connections. For crowdsourcing validation, images (n=47) were graded as for TF status (normal, TF, or abnormal without TF). The images were then posted to Amazon Mechanical Turk (AMT) for grading by untrained volunteers as either "TF" or "not-TF." Each image was graded by 10 unique Turkers in all trials for $0.10 per image. The mode of AMT grades for each image was compared to the expert grade to determine AMT grading accuracy as well as sensitivity and specificity for detecting TF.
Results :
To date, over 700 images have been successfully uploaded using our system in 2 different district surveys in Tanzania. Compressing images locally allowed for a 95% reduction in file size with attendant reductions in upload times. All normal images (n=16) were correctly identified, and only 1 of the TF images (n=11) was graded incorrectly. 4 abnormal images without TF (n=20) were graded as “TF.” Overall sensitivity was 91% and specificity was 83%. Area under the ROC was 0.87.
Conclusions :
Remotely acquired images can be successfully uploaded to a cloud server even in regions with poor connectivity, allowing for telemedicine in previously inaccessible areas. AMT users were able to rapidly and accurately identify follicular trachoma with minimal training. Further field testing is required to determine if diagnostic characteristics are acceptable in surveys with low prevalence of disease.
This is a 2020 ARVO Annual Meeting abstract.