ESCRS - FPT08.02 - Lens Layers Descriptive Metrics Derived From Deep Learning Segmentation Of Ss-Oct Images In A Cohort Of Normal And Cataract Patients

Lens Layers Descriptive Metrics Derived From Deep Learning Segmentation Of Ss-Oct Images In A Cohort Of Normal And Cataract Patients

Published 2022 - 40th Congress of the ESCRS

Reference: FPT08.02 | Type: Free paper | DOI: 10.82333/p72g-js85

Authors: Pierre Zeboulon* 1 , Christophe Panthier 1 , Jacques Bijon 1 , Damien Gatinel 1

1Rothschild Foundation Hospital,paris,France

Purpose

First to develop a Deep Learning Model to perform segmentation of the lens layers on SS-OCT images. Second, to describe key metrics of each lens layers in a cohort of cataract and clear lens patients and meausre their correlation with age and best corrected visual accuity.

Setting

Retrospective study at the Rothschild Foundation Hospital in Paris.

Methods

For each patients, 6 radial scans of the lens were performed.
A total of 309 eyes were included (175 with clear lens and 132 with cataract) .
A deep learning model was trained to segment the lens in three layers Anterior Lens (Ant), Nucleus (Nuc) and Posterior Lens (Post) on a subset of images.
All images were then used to perform metrics calculation.
For each lens layer, we calculated 3 parameters : average pixel intensity (Density), pixel intensity standard deviation (STD) and layer central thickness (CT). All parameters were averaged over the 6 radial scans of each patient. 
We tested the Pearson correlation coefficient of each parameter with age and logMAR Best Corrected Visual Accuity (BCVA)

 

Results

Mean patient age was 46.7 years old (range 19-92).

The parameters with the highest positive correlation with age were: Nuc Density (r=0.83), Post STD (r=0.81), Ant STD(r=0.76) and Nuc CT (r=0.75).

The parameters with the highest negative correlation with logMAR BCVA were: Post STD (r=-0.66), Nuc STD (r=-0.62) and Nuc Density (r=-0.62) Nuc CT (r=-0.58) and Ant STD (r=-0.57).

Conclusions

Segmentation of high resolution SS-OCT images of the Lens allowed to calculate key meetrics for each layers. This could constitute the foundation of an objective and comprehensive cataract grading system based on SS-OCT.