Abstract
Purpose :
To implement a fully automatic computer-based malarial retinopathy (MR) screening system as a tool to improve the diagnostic accuracy of cerebral malaria (CM).
Methods :
MR is a retinal manifestation of CM. It presents as a series of highly specific (> 95%) retinal lesions that can confirm the presence of CM and reduce false-positive diagnosis of CM. We developed an artificial intelligence system named ‘Auto-detection Software for Plasmodium Infection in Retinal Exams’ (ASPIRE), that provides automated point of care screening for MR. The system interfaces a fully automated retinal image analysis software with a handheld retinal camera, Pictor-Plus by Volk Optical. We tested a clinical prototype of ASPIRE in a malaria clinic in Malawi, Africa, where we acquired and processed retinal fundus images from patients clinically diagnosed with cerebral malaria. The ground truth for the presence or absence of MR was provided by an ophthalmologist using binocular indirect ophthalmoscopy. Sensitivity and specificity of the software in detecting MR was determined against the ground truth.
Results :
A retinal image dataset of 102 patients clinically diagnosed with cerebral malaria was utilized in this study, with N=72 patients with MR and N=30 subjects with no MR. ASPIRE software achieved a specificity of 100%, sensitivity of 63%, and positive predictive value of 1.0, against the ground truth. The user feedback on ASPIRE operation in the clinic confirmed its user-friendly functionality as a bedside tool that satisfies the clinical needs in Africa.
Conclusions :
A fully automated system for detection of MR as a tool for improving the diagnostic accuracy of cerebral malaria will aid in reducing the false positive diagnoses, resulting in appropriate and timely treatment for CM or other underlying diseases. In the next phase, we will demonstrate a fully integrated design of ASPIRE system making it easily accessible and affordable to the affected population in Africa.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.