In this retrospective study, we employed inferential statistics and multivariate logistic regression to identify several biomarkers associated with CSCR recurrence using clinical images.Imaging has become crucial aspect of clinical practice since introducing fast high-resolution machines. As a result, the volume of data to be analysed has increased, necessitating the development of new automatic analysis systems. The term’ biomarker” refers to morphological and structural changes that provide important information about the stage of a disease. The search for new biomarkers in retinal diseases has garnered significant interest in recent years. This research is driven by the need for earlier and more accurate diagnosis, improved prognosis, and the development of targeted therapies.This study is the first to address a fundamental clinical question using real-world longitudinal data to determine which patients are prone to experiencing recurrence following an initial, spontaneously resolved episode of CSCR. The recurrence of subretinal fluid is a negative prognostic factor for long-term visual outcomes, leading to alteration in the retinal pigment epithelium and Photo-Receptors20. The established standard treatment with Photodynamic Therapy (PDT) using verteporfin shows good efficacy in fluid resolution and preserving functional visual acuity11. Forecasting the likelihood of recurrence in patients facilitates optimising follow-up protocols and the timely administration of treatment interventions. This study aimed to identify optical coherence tomography (OCT)-derived parameters associated with CSCR recurrence using Discovery® software (by RetinAI) to identify and quantify various SD-OCT biomarkers.Our study identified nine predictive OCT biomarkers for CSCR recurrence (as shown in Fig. 2), including age, volume of Subretinal Fluid, volume of Intraretinal Fluid, area of Pigment Epithelium Detachment in the central area (1 mm around the fovea, based on a standard ETDRS grid), in the pericentral area (3 mm), and the peripheral area (6 mm). Additionally, the thickness of the following layers (or combinations thereof) in the central area was significant: Choriocapillaris and Choroidal Stroma, Outer Nuclear Layer, Inner Nuclear Layer and Outer Plexiform Layer. Three additional biomarkers were statistically significant: Choroidal Vascularity Index, a score estimating the disruption of the photoreceptors and Retinal Pigment Epithelium Layers, and an estimation of the area of the largest choroidal pachyvessels.Discovery® was validated on diseases including diabetic macular oedema, diabetic retinopathy, age-related macular degeneration, and vitreomacular degeneration. Although its AI modules were not explicitly trained on patients with CSCR, the accuracy of the retinal layer segmentation was confirmed by our annotator. The reliability of this segmentation in CSCR patients is further supported by recent studies25 that also achieved statistically significant results on CSCR datasets.Our findings contribute to the existing literature on CSCR. Imaging biomarkers associated with acute and chronic CSCR have been analysed using artificial intelligence methods25,26. The authors found a significant increase in subretinal fluid and the thickness of the outer and inner retinal layers when comparing patients with acute and chronic CSCR at baseline. However, our study identified parameters independent of CSCR type, which may better reflect real-world practice.These results align with those reported in the literature. Similar to investigations for diseases such as age-related macular degeneration and diabetic macular oedema27,28,29,30,31,32, recent studies have reported the predictive role of certain retinal biomarkers for CSCR17,33,34,35.We will now discuss each of the nine predictive biomarkers in order of their importance, as determined by the effect size (positive or negative) of each corresponding parameter in our multivariate logistic regression model applied to longitudinal data (as shown in Fig. 3). This approach provides a comprehensive overview of the relative significance of each biomarker in predicting CSCR recurrence, shedding light on their potential roles and interactions in the disease process. The strongest negative effect size was associated with Outer Nuclear Layer (ONL) thickness, suggesting that a thicker ONL may be protective against CSCR recurrence. This aligns with the notion that a thicker ONL, indicating healthier photoreceptor layers, may contribute to better retinal integrity and reduce the risk of recurrence. In the literature, ONL thinning in the fovea is positively associated with disease duration prior to treatment, suggesting that prolonged disease limits the potential for foveal ONL recovery35. Additionally, ONL thickness has been correlated with the duration of symptoms36. The positive effect size of our score for the disruption of the photoreceptors and Retinal Pigment Epithelium Layers, indicates its predictive value for recurrence. This finding is consistent with studies linking photoreceptor layer disruption with chronicity and poorer outcomes in CSCR37. The positive effect size for volume of Subretinal Fluid aligns with recent studies suggesting its role as a potential biomarker differentiating acute CSCR from chronic forms25. The positive effect size for Intraretinal Fluid indicates an association with recurrence which is novel in the literature. Pigment Epithelium Detachment (PED) showed a positive effect size in our analysis. Other studies have demonstrated that larger PED is strongly associated with persistent CSCR38, supporting our findings that PED may be a predictor of CSCR recurrence. The positive effect size for age is consistent with studies that associate older age with a higher likelihood of persistent or recurrent CSCR38. This is likely due to the cumulative impact of age-related changes in the retina and choroid. The negative effect size of a higher Choroidal Vascularity Index (CVI) aligns with literature reports that eyes with active CSCR have significantly higher CVI compared to those with resolved CSCR and healthy eyes39. The positive effect size for the thickness of the choriocapillaris and choroidal stroma is consistent with reported findings that thicker choroidal layers are typically observed in the acute stages of CSCR40, whereas persistent CSCR episodes are characterised by a thinner choroid38. The negative effect size for the combined thickness of the Inner Nuclear Layer and Outer Plexiform Layer suggests an inverse relationship with the likelihood of CSCR recurrence. This finding aligns with recent studies that highlight its role in differentiating acute from chronic forms of CSCR25.Our longitudinal analysis, extended these findings by including additional biomarkers specific to CSCR disease recurrence.Defining and validating predictive biomarkers for CSRC recurrence is crucial for several reasons in daily clinical practice. First, predictive biomarkers may enable the early diagnosis of individuals at risk of CSCR recurrence. Early diagnosis allows timely intervention and management, potentially preventing severe complications and preserving vision. Additionally, biomarkers can help tailor treatment strategies for individual patient profiles. Clinicians can offer customised treatment plans by identifying patients who are most likely to relapse.Furthermore, understanding the likelihood of recurrence allows physicians to set realistic expectations and plans for long-term management. By identifying high-risk patients, they can optimise the use of resources by defining appropriate follow-up based on the likelihood of recurrence. Biomarkers also contribute to advancing research and development by enhancing our understanding of the underlying mechanisms and risk factors associated with CSCR recurrence. This could lead to the development of more effective therapies and preventive strategies.Despite the promising results linking OCT-derived biomarkers to the future recurrence of CSCR, our study has several limitations that warrant discussion.One of the primary challenges was the variability in the number of visits per patient, particularly pronounced in chronic or recurrent CSCR cases, as these patients are examined more frequently and over longer periods. Our dataset also demonstrated significant variability in subretinal fluid volumes. Volumes ranged from minimal (approximately 10 nL) to significant (up to 10,000 nL), complicating the identification and categorisation of CSCR episodes. Although a CSCR episode is generally defined as an accumulation of subretinal fluid followed by its resolution, diverse manifestations based on the individual fluid ranges make distinguishing between multiple occurrences and a single episode challenging. This complexity underscores the challenges faced by medical professionals in accurately labelling the episodes.A particularly significant limitation of our study was the binary labelling system used to classify CSCR cases as either recurrent or non-recurrent. This classification framework, while straightforward and appropriate given the size of our dataset, requires revision to handle the inherent complexities and ambiguities of certain patient cases. In particular, the accurate categorisation of chronic patients, characterised by fluctuating subretinal fluid levels, was challenging. Applying our labelling system led to chronic patients being labelled as recurrent cases despite the nuances in their conditions, suggesting a need for more complex categorisation. This subgroup, characterised by persistently elevated subretinal fluid volumes, likely represents a distinct clinical picture. In some cases, patients with chronic diseases were excluded from the analysis. Their exclusion, while necessary for the integrity of our current analysis, highlights the limitations of a binary classification system, which future studies could address through multi-label annotation.An intrinsic limitation of our study is that some patients classified as non-recurrent, based on the data available at the time of collection, may have experienced a CSCR episode after data collection. The potential for future occurrences not captured in our dataset underscores the dynamic nature of CSCR and the challenges inherent in predicting its course.Finally, our approach does not explicitly account for the inherent time dimension in the data. Leveraging methods that consider the evolution of biomarkers over time may improve the prediction of CSCR recurrence and pave the way for translational medical applications.Our analysis revealed correlations between several biomarkers and CSCR recurrence. Applying multivariate logistic regression to all biomarkers explained 8.7% of the variance among patients. While these results are useful in elucidating each biomarker’s importance, they underscore our model’s limited capacity to predict CSCR recurrence automatically. We chose logistic regression for its simplicity and familiarity to medical professionals, aligning with our goal of making our work accessible for understanding the importance of OCT-derived biomarkers in predicting the recurrence of central serous chorioretinopathy. Although more advanced models based on Deep Learning could enhance prediction by analysing images directly rather than using extracted biomarkers, they would not have provided insights into biomarker importance.
