Comprehensive Summary
This study investigates the development of a deep learning (DL) framework for detecting and quantifying cystoid fluid collections (CFC) in spectral-domain OCT images of patients with X-linked retinoschisis (XLRS). The researchers trained a U-Net–based model using 112 OCT volumes from the RETOUCH challenge and externally validated it with 37 SD-OCT scans from 20 XLRS patients. Model performance was evaluated using Dice coefficients and intraclass correlation coefficients (ICC), yielding mean Dice scores of 0.886 on randomly sampled B-scans and 0.936 on central B-scans, comparable to those of human graders. The discussion emphasized the model’s ability to provide accurate and automated quantification but noted systematic overestimation of segmentation regions as a limitation requiring refinement.
Outcomes and Implications
This research is clinically significant because precise OCT-based fluid quantification is essential for monitoring disease progression and evaluating treatment outcomes in XLRS. The DL model shows potential to reduce the workload of clinicians, standardize image analysis, and expand access to reliable retinal disease monitoring. However, the authors note that further validation in diverse patient populations, regulatory approval, and integration into healthcare workflows are necessary before clinical deployment. With these steps addressed, widespread implementation could realistically occur within the next three to five years.