PURPOSE: To describe in vivo confocal microscopic features of Pythium insidiosum in patients with Pythium keratitis and compare with those observed in fungal keratitis. METHOD: We collected in vivo confocal images of the cornea from patients with microbiologically confirmed Pythium and fungal keratitis, analysing five putative distinguishing features: filament width (broad or thin), granularity within the filament (present or absent), filament continuity or traceability, the presence or absence of loops, and the double track sign. Three masked observers were shown images with concealed identities and tasked with detecting Pythium filaments. After initial assessment and training, their detection rates were calculated and compared before and after training. We did perform imageJ (Open Source software project Fiji) analysis of all the images for objectively assessment. RESULTS: Sixty confocal images of Pythium (n=32,15 patients) and fungal (n=28,12 patients) keratitis were analysed. The continuity of filaments and the presence of loops emerged as strong predictors of Pythium, with adjusted odds ratios (OR) of 18.1 and 19.29, respectively, based on multivariate logistic regression and decision tree splits. Pre-training accuracy was 0.51, 0.52, and 0.56, but post-training (95% CI) improved to 0.75 (0.62-0.85), 0.80 (0.67-0.89), and 0.86 (0.75-0.94). Correct identification rates for Pythium were 27, 28, and 29 (84-89%) out of 32, and for fungus were 16, 21, and 24 (57.4-85.7%) out of 28 images with sensitivity and specificity ranging from 70.7-87.5% and 80-85% respectively. ImageJ analysis revealed a significant difference between Pythium and fungal filaments in both width (9.30±1.21 μ vs. 6.20±0.88 μ, p<
0.001) and branching angle (83.92±13.57° vs. 55.10±6.03°, p<
0.001). CONCLUSIONS: Based on our analysis, these features may be indicative of Pythium and could serve as a helpful reference for future prospective studies. However, further large scale studies and validation are needed to strengthen these observations.