Photo Credit: Sruilk

The following is a summary of “Thickness Speed Progression Index (TSPI): Machine Learning Approach for Keratoconus Detection,” published in the November 2024 issue of Ophthalmology by Awwad et al. 

Researchers conducted a retrospective study to develop and validate a pachymetry-based machine learning index for distinguishing between keratoconus, keratoconus suspect, and normal corneas.  

They included 349 patients’ eyes, categorized into normal, frank keratoconus (KC), and keratoconus suspect (KCS) corneas. The KCS group comprised topographically/tomographically normal (TNF) and borderline fellow eyes (TBF) from patients with asymmetric keratoconus, 6 parameters, derived from the Corneal Thickness Progression map using the Galilei Dual Scheimpflug-Placido system, were inserted into a machine learning algorithm to generate the Thickness Speed Progression Index (TSPI). The model was trained using 5-fold cross-validation and a random search over 7 machine learning algorithms to determine the optimal model and hyperparameters.  

The results showed 33 normal eyes, 141 KC eyes, and 75 KCS eyes, including 34 TNF and 41 TBF, were included. In experiment 1 (normal and KC), the best model, Random Forest, achieved 100% accuracy and an AUROC of 1.00 for both normal and KC groups. In experiment 2 (normal, KCS, and KC), the model reached an overall accuracy of 91%, with AUROC values of 0.93, 0.83, and 0.99 for normal, KCS, and KC corneas, respectively. In experiment 3 (normal, TNF, TBF, and KC), the model achieved 87% accuracy, with AUROC values of 0.91, 0.60, 0.77, and 0.94 for normal, TNF, TBF, and KC corneas, respectively.  

Investigators concluded that machine learning algorithms like the TSPI, utilizing solely pachymetry data, can effectively differentiate normal corneas from KC or suspected KC with a satisfactory level of accuracy.   

Source: ajo.com/article/S0002-9394(24)00540-3/abstract