Colorimetric Image Analysis in Detection of Leukocoria from Retinoblastoma in Snapshots Taken by Standard Digital Photography

Abstract

Purpose: Early diagnosis and treatment of retinoblastoma are extremely important but remain challenging. Screening programs that use “red reflex” testing to detect leukocoria have improved the timing of diagnosis and treatment outcomes. Early detection of leukocoria via amateur photography may facilitate earlier diagnosis. The purpose of this study is to determine: (i) whether leukocoria can be reliably detected with recreational flash photography during early retinoblastoma, and (ii) whether its detection correlates with disease progression and treatment.

Methods: 7377 consecutive, recreational digital photographs of one patient with bilateral retinoblastoma were collected and analyzed over three years (1068 days) from birth, to retinoblastoma diagnosis, through treatment (systemic chemotherapy, enucleation of one eye and laser, cryotherapy and proton radiation therapy in the fellow eye), and remission. For control, 305 pupils of 19 healthy children present in the same photographs were analyzed. Images were converted to JPEG format, each pupil cropped, total pixel count determined using Adobe Photoshop, and retrospectively analyzed for the presence of leukocoria, typically defined by a pupil Value (brightness) ≥ 0.5 in HSV color space (Hue, Saturation, Value).

Results: Leukocoria occurred in a total of 237 photographs and began at 12 days old seen in 0.5% of pictures taken per month. The frequency of occurrence increased from birth by up to 5% per month, reaching as high as 25% per month before decreasing exponentially (t1/2 = 1.1 month) to an average of 1.5% per month after successful treatment.

Conclusions: Leukocoria can be detected in the early stages of retinoblastoma using recreational digital photography. Frequency of leukocoria occurrence in digital photographs correlated with disease progression and treatment. Development of leukocoria detection tools based on digital photography (e.g., leukocoria detection software) might improve the timeliness and accuracy of retinoblastoma diagnoses.

Keywords: 703 retinoblastoma • 550 imaging/image analysis: clinical • 688 retina