Выявление кератоконуса на основе видео динамической деформации роговицы с использованием глубокого обучения

Keratoconus Detection-based on Dynamic Corneal Deformation Videos Using Deep Learning.

Keratoconus (KC) is a progressive ectatic disease characterized by corneal steepening and thinning due to deficits of structural integrity inducing irregular astigmatism and myopia with deterioration of visual acuity and vision-specific quality of life.^1,2 Early diagnosis of KC is of foremost importance to halt disease progression.³ Corneal topography or tomography systems allow assessment of alteration of corneal shape and assist in KC classification and monitoring.⁴ However, these instruments cannot measure the mechanical stability of the cornea, which is thought to be the primary abnormality that triggers stromal thinning and steepening.⁵ Therefore, the early diagnosis of KC by analyzing corneal biomechanical properties has become a research hotspot aiming to measure the in vivo biomechanical properties of the cornea before the onset of topographic or tomographic changes.^5,6 The Corvis ST (Oculus Optikgeräte GmbH) was introduced as a noncontact tonometer that records the ocular deformation response induced by a constantly defined air puff using an ultrahigh-speed Scheimpflug camera. The captured images are analyzed by the Corvis ST software to produce an estimate of intraocular pressure and several dynamic corneal response (DCR) parameters that provide a detailed assessment of corneal biomechanical properties. These parameters can be used to separate normal (N) from ectatic corneas.⁶, ⁷, ⁸, ⁹ A video function uses these images to release a slow-motion video of the corneal deformation reaction in response to the air pulse (4330 frames/second.). This highly precise video permits a viewable biomechanical analysis of the cornea.¹⁰