Objective assessment of corneal transparency in the clinical setting: Identification and correction of acquisition artifacts in SD-OCT images

Abstract

Purpose : To develop an automated algorithm for clinical spectral domain OCT (SD-OCT) images, capable of correcting hyperreflective artifacts associated with the instrumental configuration and the patient’s eye position.

Methods : The automated pre-processing algorithm (Python Software Foundation, v2.7.4) developed standardizes raw images (flattening, segmentation, normalization) and computes a correction mask from each OCT image. Our approach is based on the Principal Component Analysis (PCA) of the in-depth stromal OCT signal, which enables the statistical computation of the device’s point spread function and permits its compensation (Fig.1). The mean stromal intensity depth profile is then extracted and analyzed via our previously developed method designed to quantify the photon mean free path in the stroma (R Core Team, v3.6.3) as objective measure of corneal transparency. We tested our method and associated algorithm on clinical SD-OCT images acquired with an RTVue-100 OCT device (Optovue, Inc.). n=37 normal corneas (aged 33 ± 8 years) were studied (2 images per eye). Patients were chosen according to the inclusion criterion of pre-refractive surgery, implying clear and healthy corneas.

Results : Results are shown in Fig.2. Boxplots depict the photon mean free path values (l_s) obtained by our image-analysis method for each eye imaged. Typical l_s values are on the order of 500 – 1000 µm. We obtain a log-normal distribution of results for data up to the third quartile with a 6.9 % coefficient of variation: <log(ls)> = 6.1 ± 0.4 dB, Shapiro-Wilk test p-value = 0.5.

Conclusions : Our pre-processing algorithm is capable of robust detection and correction of clinical SD-OCT image artifacts. It enables to determine numerical values of scattering mean free path in vivo using unmodified clinical diagnostic devices. Future steps will include a correlation of our method with clinical data (e.g., corneal densitometry) obtained by other methods (Bland-Altman comparison), as well as with other clinical measures (e.g., visual acuity).

This is a 2021 ARVO Annual Meeting abstract.

 

Illustration of artifact detection and correction in posterior stroma.