Abstract
Purpose :
To develop a fully automated and low-cost system for malarial retinopathy (MR) screening to improve the diagnostic accuracy of cerebral malaria (CM).
Methods :
Cerebral malaria (CM) remains the major killer of children in Africa; however, up to 25% of these deaths are due to misdiagnosis of CM. MR is a highly specific retinal manifestation of CM that can improve the diagnostic accuracy of CM. We developed an artificial intelligence system, ASPIRE, that integrates an MR detection software with a low-cost retinal camera, iNview. The software model was trained using Deep Learning techniques and tested on retinal fundus image data of N=125 patients clinically diagnosed with CM in a malaria clinic in Malawi, Africa. The ground truth for the presence or absence of MR in retinal images was provided by a certified retinal reader. The users recorded their feedback on ASPIRE’s usability in a malaria clinic.
Results :
The dataset of 125 patients included N=84 patients with MR and N=41 patients with no MR. The MR detection model provided a specificity of 100% and sensitivity of 90% with an AUC of 0.98. The users noted ASPIRE as a user-friendly, ergonomic, and bedside diagnostic tool, suitable for the clinical flow in Africa.
Conclusions :
ASPIRE will improve the diagnostic accuracy for a fatal neurological complication of cerebral malaria (CM); which will save the lives of hundreds of thousands of children in Sub-Saharan Africa. In the next phase, ASPIRE functionality will be refined to make it easily accessible and affordable to the targeted population in Africa and validated in a large clinical study.
This is a 2020 ARVO Annual Meeting abstract.