Predicting Manifest Astigmatism in Keratoconus Patients: a Machine Learning Approach Using Corneal Topography.
试验速览
- 阶段
- 不适用
- 状态
- 进行中（未招募）
- 发起方
- 入组人数
- 800
- 试验地点
- 1
- 主要终点
- The root mean squared error (RMSE) of the machine learning model's predictions for the patient's manifest astigmatism.
研究概览
简要总结
The cornea of the human eye has several functions. It is transparent to allow light into the eye and its shape focuses the incoming light onto the retina. The cornea is responsible for two-thirds of this focusing, while the human lens accounts for the remaining third.
Keratoconus is a condition with onset in the second to third decades of life, where the cornea warps into an irregular shape. This irregularity reduces vision due to blurring of the image focused on the retina. Only partial improvement is achievable with traditional glasses because the corneal shape becomes irregular.
The glasses prescriptions of patients with keratoconus often differ significantly from the measurements obtained from the cornea in a clinical setting. The predictability of the magnitude and variability of this disparity is poorly understood. As a result, determining the optimal glasses prescription for achieving the best vision correction often involves a time-consuming trial-and-error approach. Improved predictability could reduce the time required to identify the optimal glasses prescription, thereby increasing productivity. For surgical patients, better predictability would enable surgeons to select lenses that provide superior vision outcomes after treatment. In the optometry clinic at the West of England Eye Unit, a database of over 800 patients with glasses prescriptions and corresponding corneal scans has been compiled by the investigators. This is a sufficient dataset to train and assess the prediction accuracy of machine learning models (AI) of glasses measurements using corneal scan parameters as predictor variables.
详细描述
Keratoconus is one of the leading causes of visual impairment worldwide. Its main therapeutic options include corneal cross-linking, intracorneal ring segments, penetrating keratoplasty, and lamellar keratoplasty.
Keratoconus is a progressive condition with onset in childhood, resulting in increasing irregularity of corneal shape. The cornea's role is to refract incoming light to focus it onto the retina to produce a sharp image. Irregularity of the cornea degrades this image. Keratoconus develops throughout the second and third decades of life and usually stabilises in the fourth decade. It leads to progressive vision loss. When detected early, progression can be halted or stabilised with collagen cross-linking treatment. Once the condition progresses to impair the quality of vision, rigid contact lenses or surgical procedures are often required for correction. Spectacles and soft contact lenses are frequently inadequate due to the cornea's irregular shape in keratoconus.
Surgical approaches to keratoconus that do not involve corneal transplantation include implantable collamer lenses (ICL) or toric intraocular lenses (tIOL). There is little published evidence on the use of these approaches, and no consensus exists regarding the extent of astigmatism to correct or the proper axis of astigmatism to select.
In addition to lower-order aberrations (such as those correctable by spectacles, including short-sightedness, long-sightedness, and astigmatism), higher-order aberrations are present, which cannot be corrected with glasses. Due to irregular corneal astigmatism and higher-order aberrations, accurate refraction is often challenging in patients with keratoconus. Eyes with keratoconus exhibit approximately five to six times more higher-order aberrations than normal eyes, particularly vertical coma. The manifest refractive cylinder (the astigmatism component in a glasses prescription) includes both regular and irregular astigmatism. Additionally, the axis of astigmatism in the prescription often differs significantly from the topometric axis derived from three-dimensional corneal imaging. In managing astigmatism for suitable patients, such as those receiving implantable collamer lenses (ICL) or toric intraocular lenses (tIOL), it is crucial to better understand the shape factors contributing to manifest refraction. Improved understanding can help neutralise post-operative refraction more effectively and achieve greater predictability. For non-surgical patients, determining the optimal glasses prescription to achieve the best vision correction often involves a time-consuming trial-and-error process. Improved predictability could reduce the time required to identify the optimal prescription, thereby enhancing productivity and patient care.
The aim of this study is to use data already collected from more than 800 patients to compare topographic elements (three-dimensional modelling of corneal shape), such as anterior and posterior curvatures and higher-order aberrations, to subjective manifest astigmatism (the amount of astigmatism in the glasses prescription) in keratoconic eyes. The impact of keratoconus severity on disparities between these methods will also be evaluated.
研究设计
- 研究类型
- Observational
- 观察模型
- Case Only
- 时间视角
- Retrospective
入排标准
- 年龄范围
- 16 Years 至 —（Child, Adult, Older Adult）
- 性别
- All
- 接受健康志愿者
- 否
入选标准
- •All patients with a diagnosis of keratoconus who have had topography performed at the West of England Eye Unit with contemporaneous manifest refraction.
排除标准
- •Poor quality topography scans.
- •History of ocular surgery.
- •Age <16 years.
结局指标
主要结局
The root mean squared error (RMSE) of the machine learning model's predictions for the patient's manifest astigmatism.
时间窗: Baseline
The primary outcome measure will be derived from the machine learning regression analysis, with demographic and keratometry values serving as predictor variables (X) and manifest astigmatism as the outcome variable (Y). For each patient, the earliest Pentacam scan recorded during the study period, for which a manifest astigmatism measurement exists on the same day, will be used.
次要结局
- Shapley values for each predictor variable in the best-performing machine learning model.(Baseline)
研究者
Harry Roberts
Principal Investigator
Royal Devon and Exeter NHS Foundation Trust