Optimization of diagnostics of retinopathy of prematurity stages based on the integration of clinical data using the Key to Diagnosis I software

Abstract

Despite the improvement of algorithms of preterm infants’ management, methods for predicting, diagnosing and treating ROP remains a vital issue. Purpose: to improve the diagnostics of retinopathy of prematurity based on the assessment of vascular system configuration, using Key to Diagnosis I software. Material and methods. 279 patients with ROP were divided into 6 groups: group 1 included 152 patients (304 eyes) with stage I; group 2 — 45patients (90 eyes) with stage II; group 3 — 8patients (12 eyes) with stage III; group 4 — 7 patients (8 eyes) with stage IVA; group 5 — 7 patients (14 eyes) with posterior aggressive ROP; control group 6 — 60patients (120 eyes) diagnosed with immature retina who have no ROP signs. 28 eyes were analyzed using wide-field imaging, while 400 eyes were analyzed by separate images. The presence o f“mute” zones, macula localization, traction index of the macular zone (Tm), zone and span of pathological changes, fractal dimension (Df) and complexity of vascular system (СVS) were assessed on automatically created wide-field images, obtained by Ret-Cam Shuttle. Results. We revealed strong correlation between Df and stages (p = 0.85, p = 0.01); moderate negative correlation of Тm and stages (p = 0.62, p = 0.01), except for posterior aggressive ROP; strong positive correlation between CVS and stages ( p = 0.91, p = 0.001). Diagnostic modules of the software have been developed to create wide-field fundus imaging in infants, localize the macula as a marker for morphometry, and isolate the vascular system using deep convolutional neural networks. Conclusions. The developed algorithm for multivari­ate analysis of the retinal vascular system reduces the risks of subjective assessment of retinal changes.