Looks like they are happy with what has been released so far...lets hope we can get restarted soon...
Keywords: DR (Diabetic Retinopathy), NPDR (None Proliferative Diabetic Retinopathy) fundus, PDR (Proliferative Diabetic Retinopathy), CARA (Computer Automatic Retinal Analysis), AI Aartificial Intelligence) and ETDRS (Early Treatment of Diabetic Retinopathy Study).
Background:
Diabetes is a major public health problem, associated with high rates of morbidity and mortality in any human community. It is well understood that the burden of diabetes is attributed to chronic progressive damage in major end-organs. Therefore, the Diabetic retinopathy (DR) is a serious and frequent complication and the leading cause of blindness in adults’ population.
Aims: The overall purpose of the present project is the harmonization and optimization of ISSSTE structures in order to reduce wait times, improve timeliness and maximize management of patients with diabetic condition in Mexico City. CARA telemedicine platform based on AI program from Diagnos.inc was used for mass screening, triage and management of the different follow up visit.
Methods: 79000 Patients with diabetes and prediabetes condition were screened (32%male, 67% female and 1% other gender not specified), a mean age (59.8 ±13.4 Years), all the participants underwent automatic screening by CARA Tele-Retinal system, the picture acquisitions were performed by a qualified professional in a darkened room acquired by an nonmydriatic fundus camera, the median time for screening for DR was (10mn ±2). The picture was taken from each eye (OD and OS) with the macula in the center (45 degree field of view) according to clinical gold standard protocol and Early Treatment of Diabetic Retinopathy (ETDRS) criteria, semi-automatic classification was performed by CARA algorithm DR0 (normal fundus), DR1 (Mild NPDR), DR2 (Moderate NPDR), DR3 (Severe NPDR) and DR4 (PDR). The receiver operator characteristic curve (AUC) was used to assess the improvement in detection of any DR by automatic detection and classification mode using CARA algorithm compared to masked trained optometrist mode.
Results: 53.04% of the patients had history of diabetes condition, 46.96% of the patients had no history of diabetes, while 37.7% of the patients were found with no DR, DR was automatically detected by CARA algorithm in 61.5% of the patients, those with DR1 condition were (36.8%), DR2 (11.5%), DR3 (8.9%) and DR4 condition were (4.3%). Therefore, 0.8% of the patients there had unclassified images of retinal fundus, while 7% of the patients detected with severe condition of DR were referred urgently to the specialist for treatment. The sensitivity of detecting any DR by automatic grading system (CARA algorithm) was 79.8% (95% CI, 72.1% -87.5%), the specificity was 87.7% (95% CI, 82.3%-93.0%) with an AUC of 88.5%.
Conclusion: Telemedicine program through AI application was successfully implemented and executed on a large scale population with significant improvements, allowing faster and comfortable DR detection programs for larger populati.on with high efficiency in the safety way. These results, highlights the strength and simplicity of such an automatic telemedicine program with the potential to maximize access and standardized monitoring during the fellow up visits with best cost clinical effectiveness and costeffectiveness.
References: Criterios de Operacin del Escalamiento del Manejo Integral de Diabetes por Etapas MIDE. ISSSSTE. Mexico, 2016. Early Treatment Diabetic Retinopathy Study Research Group. Early treatment diabetic retinopathy study design and baseline patient characteristics. ETDRS report number 7. Ophthalmology 1991;98:741– 756.
Observatorio de Diabetes y Enfermedades Crnicas “ODEC”. ISSSTE Mexico 2016-2017.