Real-Time, Clinician-Free Detection of Staphyloma Presence and Apex Location in a Cohort of Highly Myopic Eyes With an Ultrasound-Based Algorithm

Abstract

Purpose : Ultrasound is a commonly available tool to assess the properties of the posterior eye, allowing us to characterize morphological changes. Here, the present study investigates the use of curvature in a simple algorithm to achieve automated (clinician-free) staphyloma detection, as well as pinpoint the location of staphyloma apexes— locations known to be prone to pathologic change.

Methods : 46 individuals (emmetropic, highly myopic (HM) or pathologic myopia) were enrolled in this study (axial length (AxL) range: 22.3-39.3mm) yielded 130 images in total. 10 MHz US B-scan images were acquired while subjects fixated in primary gaze. Eyes were clinically classified into two groups – a non-staphyloma group (composed of eyes that were either emmetropic or HM without staphyloma) and a staphyloma group. On each US image, an intensity-based segmentation algorithm automatically tracked the posterior eyewall. Local curvature (K) of the posterior eyewall and distance (L) of the posterior eyewall to the US transducer were calculated. The location of the staphyloma apex was also automatically estimated. The area under the receiver operator characteristic (AUROC) curve was used to evaluate the diagnostic ability of eight local statistics derived from K, L and AxL. The performance of binary classification (i.e. presence or absence of staphyloma) was assessed at an optimal cut-off point and compared with the performance of junior clinicians.

Results : A fully-automated algorithm was able to detect the posterior eyewall and quantify the non-uniformity of the posterior eye shape with a good classification performance of AUROC > 0.70 for most parameters derived from local curvature (K). The best classifier (the combination of AxL, standard deviation of K, and the standard deviation of L) yielded a diagnostic validation performance of 0.897, which was comparable to the diagnostic performance of junior clinicians. Our method localized the staphyloma apex with an average error of 1.35+/-1.34 mm.

Conclusions : Our fully-automated method enables staphyloma detection with performance comparable to that of junior clinicians. Combined with real-time data acquisition capabilities of US, this method has the potential to be employed as a screening tool for clinician-free in-vivo staphyloma detection.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.