Major depressive disorder (MDD) is a psychiatric disorder characterized primarily by depressed mood and anhedonia. With a lifetime prevalence of 16%, MDD is one of the leading causes of disability worldwide (1, 2). Currently, the diagnosis of MDD mainly relies on patients’ clinical symptoms and Hamilton Depression Rating Scale (HDRS). Although their diagnostic value in clinical practice should be recognized, relevant limitations should not be ignored, particularly the unavoidable subjectivity and the insufficient reliability, which leads to the insufficient sensitivity and specificity for early-stage MDD screening. Therefore, developing objective and quantifiable diagnostic indicators is essential to enhance the diagnosis of MDD.

The pathophysiological mechanisms of MDD are complex and closely related to the genetic, environmental, neuroendocrine, cardiovascular, and psychosocial factors (3–6). In addition, the important role of certain biomarkers in the pathogenesis of MDD has been confirmed by several studies, such as 5-hydroxytryptamine, cortisol, and brain-derived neurotrophic factor (7–9). Growing evidence suggests that cerebral microvascular dysfunction may facilitate the development of MDD by inducing chronic ischemia in brain tissue (10). Clinically, positron emission tomography and functional magnetic resonance imaging (MRI) have become a necessary addition to the diagnosis and monitoring of MDD patients (11). Nevertheless, the use of these methods for population-level screening is limited by their high cost, invasiveness, long examination times, or low resolution. Therefore, after fully weighing the cost of healthcare, potential physical harm and overdiagnosis, finding a simple, effective, and fast method to identify neurological abnormality and cognitive profiles is essential for comprehensive and personalizing management, optimizing treatment strategies, saving medical resources, and preventing complications in MDD patients.

In recent years, the relationship between MDD and neurodegenerative diseases has received great research interest. In addition, several studies have revealed that patients with MDD have neurodegenerative changes in the central nervous system, with functional and structural damage in multiple regions of the brain (11–14). As an extracranial extension of the cerebral nerve, the retina provides an easily observable window for monitoring neurodegenerative disease. The retina shares similar physiological and anatomical characteristics with the cerebrum, and the alterations of retinal structural and microvascular parameters may reflect similar variation of those in the cerebral or other end organs (15–19). Recently, thinner nerve fiber layer thickness and enlargement of retinal venules have been reported in patients with mental disorders (20–22). However, traditional methods of measuring retinal microvasculature, such as fundus photography and color Doppler imaging, have the limitation of inconvenient operation and low reliability, which results in the inability of these methods to be widely used in MDD.

Optical coherence tomography angiography (OCTA) provides a unique, quantitative, non-invasive method of assessing the retinal structural and microvascular system, and has been one of the fastest growing and most prospective non-invasive indicators in the field of ophthalmology, and even extra-ophthalmology (23). Interestingly, the use of OCTA as the indicator for neurodegenerative diseases, such as cerebrovascular disease, cognitive impairment and schizophrenia (SP), have been widely reported in recent years (24–27). However, the relationship between MDD and the retinal vasculature remains controversial.

To address these questions, our study enrolled 78 patients with MDD and 47 healthy controls (HCs) to quantitatively evaluate the retinal vessel density and structure changes in patients with MDD by OCTA, and then screen out more reliable retinal indicators for MDD patients through a strict research design. Furthermore, a strong, innovative retina-based diagnostic model for early detection of MDD has been further developed and verified.

The retina is the only visualized neurovascular tissue in the body, providing an easy-to-see window for monitoring systemic neurovascular conditions (30, 31). OCTA is a novel, non-invasive and quantitative retinal imaging tool of which use has been extended from early screening and monitoring of fundus diseases to systemic diseases (32–35). OCTA-base retinal parameters, such as retinal vessel density and thickness, are expected to be one of the most promising and competitive non-invasive indicators in clinical practice.

As a neurodegenerative disease, the diagnosis of MDD mainly relies on the patient’s subjective assessment and lacks quantifiable objective indicators. Although previous studies have demonstrated the structural alterations of the retina in MDD patients, such as thinner RNFLT (36), reduced minimum ganglion cell-inner plexiform layer and poor visual function (37), a highly reliable retinal-related indicators for screening of early-stage MDD remains to be identified for the following reasons. First, previous research indicators on the retina of MDD patients are relatively single, mostly focusing on retinal thickness, and a comprehensive and systematic analysis of the retina is lacking. Furthermore, most of the previous studies had relatively small sample sizes; therefore, the relevant results were not very reliable. Finally, the accuracy of individual retinal parameters is generally limited, hinting at the need to build diagnostic models based on multiple retinal parameters to improve screening capabilities.

Given the above considerations, we performed a comprehensive comparative analysis of retinal parameters between MDD patients and HCs. We found that MDD patients had lower retinal vessel density, thinner SFCT, and poorer visual function compared to HCs. Recently, abnormalities in retinal vessel density and structure have also been demonstrated in psychiatric disorders, such as BD and SP (26, 27, 38–40). As for MDD, although there is no definite evidence directly proving the association between MDD and retinal vessel density, retinal vascular dysfunction and endothelial damage in MDD patients have been confirmed in previous studies (41, 42). Furthermore, the inflammatory hypothesis of MDD may also partially explain our findings. Interestingly, increased inflammatory cells and acute phase proteins have been demonstrated in the cerebrospinal fluid and peripheral blood of MDD patients (43, 44). Inflammatory processes have been demonstrated in several psychiatric disorders, such as SP and depression (45). The inflammatory process may trigger microvascular thrombosis and endothelial inflammation, which may lead to a reduced retinal blood supply and blood flow velocity (46). In addition, compared to other tissues with collateral vessels, the retinal plexus consists of terminal arteries without the anastomotic connections and is more affected by hypoperfusion (47). Retinal and choroidal vascular abnormalities in MDD patients may lead to insufficient supply of nutrients to the retina/choroid, resulting in choroidal thickness changes and visual field damage. Thus, the above factors may contribute to explain the underlying pathophysiological mechanisms of retinal morphological functional impairment in MDD patients in our study.

Moreover, to enhance the accuracy of the above retinal parameters in MDD screening, we developed a new retina-based diagnostic score by appropriate statistical tools, including SCP-VD, DCP-VD, outer-circle RPC-VD, SFCT, MD, and PSD. The results strongly indicated that our model may have powerful diagnostic capabilities for early-stage MDD, which was better than each of the individual retinal parameters. More importantly, based on the diagnostic score, a corresponding visual nomogram was established to enhance its clinical applicability. Consistent with our findings, alterations of above six retinal parameters in psychiatric disorders and their potential clinical roles or pathophysiological impact were more or less revealed in earlier researches (27, 38, 39). To some extent, the retinal parameters mentioned above may be closely related to the development of MDD, which suggests that the corresponding morphological and functional changes in the retina may mirror the pathophysiological alterations of MDD in a timely and reasonable manner.

Given the promising performance of the retina-based model score for MDD diagnosis, this model may provide a valuable guidance for early screening and efficacy assessment of MDD. On the one hand, some MDD patients usually conceal their condition from others out of fear of public rejection, thus further prolonging their suffering. For the screening of these high-risk groups, relying on subjective symptoms and depression scale scores alone is often prone to underdiagnosis and misdiagnosis, which may result in serious consequences. Therefore, the use of retina-related artificial intelligence models for early screening of people at high risk of MDD and targeted specialty evaluation and treatment of patients with suspected MDD may be helpful for reducing healthcare costs and self-harm or suicide due to underdiagnosis. On the other hand, with further in-depth research, retina-base model scoring of MDD patients at different stages of the treatment process may be useful in assessing efficacy, guiding treatment and predicting prognosis. However, the transition from theory to practice still faces many challenges. In the future, with the further development of artificial intelligence, the retina-based MDD diagnostic model will have a broader clinical application prospects.

There are still some limitations of this study in spite of the promising results. First, the sample size of this cross-sectional study is small, the effects of disease severity and duration on the retina have not been adequately observed. In addition, this study mainly observed patients with unmedicated depression, and the effect of antidepressant on retinal blood vessels and structures are unclear, which needs to be confirmed by a further follow-up study. Furthermore, although the diagnostic accuracy of MDD was demonstrated in our study, its capability to identify early MDD needs a further confirmation in large prospective studies and validation with external data to get additional efficacy evidence. Moreover, future studies could perform comparative analysis of retinal parameters in multiple psychiatric disorders to confirm the specificity of retinal changes in different psychiatric disorders. Finally, the direct relationship between cerebral vascular lesions on MRI and retinal vessel density using OCTA in patients with MDD requires a more systematic study to confirm.

In conclusion, based on a rigorous study design and reliable data, we identified a unique set of MDD-related retinal parameters. We found that patients with MDD have lower retinal vessel density, thinner retinal structure and poor visual function. Moreover, a credible diagnostic score based on the retina and corresponding visualized nomogram were developed by a machine learning approach. Undoubtedly, the role of a patient’s clinical symptoms and HDRS score in MDD diagnosis is irreplaceable. However, with the development of artificial intelligence technology and interdisciplinary collaboration, the retina-based model may become a promising alternative for early detection of MDD in clinical practice, thus ensuring the best timing for early detection and treatment.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by Ethics Committee of the First Affiliated Hospital of Jinan University. Written informed consent to participate in this study was provided by the participants’ legal guardian/next of kin.

JZ, YJ, XL, and LL: general study design and drafting the manuscript. XL and SL: worked on data acquisition. XL: writing original draft and writing review and editing. GY, HY, JZ, LL, SL, SM, SZ, XL, and YJ: critical revision of the manuscript. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.