Dr.Gairik Kundu

Dr. ROHIT SHETTY, Dr. KRATI GUPTA

Abstract

Aim-To develop AI model to assess local vs global progression of KC using multiple tomography parameters.Method-1518 Pentacam scan of 366 eyes analysed.Increase in Kmax was used to classify progression.Corresponding changes in other Pentacam parameters were added.3 AI models trained with increase in Kmax 0.75D (A)1.0D(B),1.25D (C)&were built using random forest.AUC, sensitivity(se),specificity(sp)&accuracy(ac) with other metrics were evaluated.Result:ModelA-AUC, se,sp,ac 0.90, 85%, 82%,83%,ModelB 0.91,86%, 82%,88%&ModelC 0.93,89%,81%,91%.All Models predicted 62% actual progression eyes had progression associated changes in other parameters also.A discordance in increase in Kmax and change in other parameters in38% eyes seen.Conclusion-The AI model identified eyes where increased Kmax matched corresponding associated changes in other parameters.These eyes had greater degree of progression&may need crosslinking earlier than remaining KC eyes,where only increase in Kmax was observed.