Towards Automating Retinoscopy for Refractive Error Diagnosis

Refractive error is the most common eye disorder and key cause behind correctable visual impairment, responsible for nearly 80% of impairment in US. can be diagnosed using multiple methods, including subjective refraction, retinoscopy, autorefractors. Although refraction gold standard, it requires cooperation from patient hence not suitable infants, young children, developmentally delayed adults. Retinoscopy an objective method that does require any input patient. However, retinoscopy a lens kit trained examiner, which limits its use mass screening. In this work, we automate by attaching smartphone to retinoscope recording retinoscopic videos with wearing custom pair paper frames. We develop video processing pipeline takes as estimates net refractive based on our proposed extension mathematical model. Our system alleviates need performed untrained examiner. clinical trial 185 eyes, achieved sensitivity 91.0% specificity 74.0% diagnosis. Moreover, mean absolute approach was 0.75±0.67D estimation compared measurements. results indicate has potential used retinoscopy-based screening tool real-world medical settings.