Forme Fruste Keratoconus Diagnostic Ability Can Improve With AI Algorithm

Forme fruste keratoconus diagnosis can be improved with the application of an artificial intelligence (AI) algorithm to a combination of noncontact tonometry and spectral domain-optical coherence tomography (SD-OCT), according to a study published in Contact Lens & Anterior Eye. 

Researchers retrospectively reviewed data from participants composing 4 age- and sex-matched cohorts, including control group participants (n=271; 48% women; mean age, 22.24 years), individuals with forme fruste keratoconus (n=84; 48% women, mean age 23.58 ±5.71 years), patients with early keratoconus (n=85; 47% women; mean age, 22.53 years), and patients with advanced keratoconus (n=159; 48% women; mean age, 21.79 years). All study participants underwent Scheimpflug tomography, SD-OCT, and air-puff tonometry. 

The team built a total of 14 AI models — 7 based on random forest and 7 based on neural networks to discriminate forme fruste keratoconus from eyes without keratoconus, and assessed area under the curve (AUC), specificity, and sensitivity to determine accuracy.   

“Although tomographic and biomechanical devices are capable of capturing different properties of the cornea, it is impractical and uneconomical to use every available device to make a diagnosis,” according to the study authors. “It is therefore important to determine the value of the added information needed for diagnosis, whether provided by a single device or a combination of three devices with different working principles.”

A combination of noncontact tonometry and SD-OCT guided by random forest AI demonstrated the best ability to detect forme fruste keratoconus (AUC, 0.902), followed by a combination of all 3 methods (AUC, 0.871). Forme fruste keratoconus detection using a single diagnostic technique was best achieved by the air puff tonometer (AUC, 0.801), the report shows. 

“[E]xisting AI methods can accurately diagnose [early keratoconus] and

[advanced keratoconus] but their diagnostic ability of [forme fruste keratoconus] is low,” according to the researchers. “Compared to using a single device, combining air-puff tonometry with Scheimpflug tomography or SD-OCT can vastly improve the diagnostic efficiency for [forme fruste keratoconus] but simultaneous use of three devices offers no additional advantages.”

Study limitations include a small selection of commercially-available noncontact tonometry and SD-OCT devices.