Detecting Structural changes in the retina of neurodegenerative patients through OCT supervised segmentation

doi:10.21203/rs.3.rs-2615323/v1

Download PDF

Article

Detecting Structural changes in the retina of neurodegenerative patients through OCT supervised segmentation

https://doi.org/10.21203/rs.3.rs-2615323/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background

To evaluate alterations of the choroid in neurodegenerative disease patients versus healthy controls using a custom supervised algorithm based on superpixel segmentation.

Methods

The prospective study protocol and procedures were approved by the Aragon Research Ethics Committee (CEICA) according to Declaration of Helsinki. Informed consent was obtained from all participants included in the study. Swept-source optical coherence tomography (OCT) B-scan images obtained using a Triton (Topcon, Japan) device were compiled. Images were included from three cohorts: multiple sclerosis (MS) patients and Parkinson disease (PD) patients diagnosed by a neurologist, and healthy subjects. 104 OCT B-scan images were processed using a custom supervised superpixel segmentation (SpS) algorithm to detect boundary limits in the choroidal layer. The algorithm groups pixels with similar structural image properties to generate clusters with similar meaningful properties. SpS automatically selects and groups the superpixels in a segmented choroidal area, computing the choroidal optical image density (COID), the total choroidal area (CA), and choroid density (CD).

Results

CA and CD were significantly reduced in the two neurodegenerative diseases (higher in PD than in MS) versus healthy subjects (p < 0.001). COID increased significantly in PD patients versus MS patients and in MS patients versus healthy controls (p < 0.001).

Conclusions

The SpS algorithm detects choroidal tissue and significant changes in choroidal images proprieties in MS and PD patients versus healthy controls. Application of the SpS algorithm to OCT images potentially acts as a non-invasive biomarker for early diagnosis of MS and PD.

Biological sciences/Computational biology and bioinformatics/Data processing

Biological sciences/Computational biology and bioinformatics/Image processing

Biological sciences/Computational biology and bioinformatics/Predictive medicine

Health sciences/Biomarkers

Health sciences/Medical research

Health sciences/Neurology

artificial intelligence

choroid

neurodegenerative disease

Parkinson Disease

Multiple Sclerosis

OCT

image processing

optical density

Ocular medium transparency makes it possible to develop new non-invasive techniques for diagnosing and monitoring neurological, vascular, and inflammatory pathologies. It allows practitioners to observe blood perfusion; to observe vascular wall integrity in hypertension, diabetes, and other diseases; to observe ocular or systemic diseases; and to observe axonal damage to the optic nerve in neurodegenerative disease. The eye is a window to the body and provides practitioners with non-invasive access to the brain, allowing them to study visual response, and to observe the structure of the ON and retinal neurons and their distribution and characteristics. [(1, 2, 3)]

Optical coherence tomography (OCT) is a technique used to diagnose ocular pathologies and to observe the retina's neuronal and vascular structures. OCT is based on interferometry between two beams of light: a reference beam and a beam that traverses the ocular media and is refracted and reflected in their structures according to changes in their refractive index. Based on this information, the OCT software creates a 3D representation of the retina, allowing practitioners to observe its structure in real time. The OCT algorithm interprets the recorded interference to determine the geometric path travelled by the examination beam and expresses it as retinal layers thickness. These structures are affected in different pathologies and the changes are detectable by OCT: the OCT software estimates if there is an alteration in each eye when comparing the records against a normalized database of healthy subjects of the same age.

Many research groups have devoted their efforts to discovering how Neurodegenerative Diseases and the associated neuronal affectation are reflected in the optic nerve and retinal nerve fiber layer (RNFL). Commercial OCT image processing focuses on the diagnosis of ocular disease; however, researchers use OCT to monitor and evaluate Neurodegenerative Disease, its progression, and even the effectiveness of treatment. Optic nerve atrophy and RNFL thinning are two findings associated with multiple sclerosis (MS). Affectation is related to the degree of severity of the disease and to the number of years since onset. Axonal loss is generally considered to be the main disability in MS patients and it is observable in the RNFL. [(4)] Retinal macular thinning and RNFL alteration in the eyes of Parkinson disease (PD) and Alzheimer disease (AD) patients have been detected by different research groups and this axonal affectation has been associated with neurological progression. [(5, 6, 7)] Alterations to perivascular spaces are a potential biomarker of AD that can be observed in the retina using OCT and/or OCT-mediated angiography. [(8)]

New OCT acquisition systems deliver high-resolution images, and subsequent software analysis of the retinal layers and choroid strengthens these findings in the study of Neurodegenerative Disease. In recent years, the principal aim has been to find a biomarker of Neurodegenerative Disease in the neuroretina or choroidal vascular tissue and to develop artificial intelligence (AI) algorithms with which to analyze images and predict or detect these pathologies. [(9, 10, 11)] These algorithms can be developed using raw data from OCT or from OCT images. Uppugunduri et al developed an automated methodology to detect separation of the Haller and Sattler layers of the choroid with the aim of quantifying the Haller layer in the choroid using OCT images. The algorithm extracts the binarized choroidal layer, identifies the choroidal vessel, and extracts large-vessel OCT B-scan images. [(12)] Analysis of the choroidal tissue of MS and AD patients using OCT software shows thinning versus healthy subjects.[(13, 14, 15)] However, the same analysis in PD patient eyes shows varying results. Garcia et al (2017) detected choroid thickening in the peripapillary zone in PD patients’ eyes versus healthy eyes after comparing parameters using SS-OCT software outputs. Satue et al (2017) made the same findings in macular zone thickness also using SS-OCT outputs. However, in 2022 the authors used angiography mediated by OCT technology and did not find any differences in vasculature dynamics and structure between groups.[(16, 17, 18)] In the other hand, Eraslan et al (2016) and Moschos et al (2017) found a decrease in choroidal thickness. Eraslan et al already include in their work image processing procedures, independent of the OCT software, they used an image-processing technique to flatten the B-scan view offered by the OCT and manually delimited the distances at different pre-established points between the Bruch membrane (BM) and the hyper-reflective sclerochoroidal junction. [(19, 20)] The aim of this study is to analyze OCT B-scan images taken from healthy subjects and neurodegenerative disease patients using a new supervised superpixel-based algorithm. Analysis focuses on detecting boundaries in choroidal tissue and structural alterations to the choroid due to neurodegeneration. This new technique is able to detect gradual changes in optical density in OCT images and to calculate the area of the choroidal layer. Our primary interest is to improve knowledge in this area, especially in patients with PD. The technique could also provide new strategies for the detection and monitoring of neurodegenerative diseases, thereby avoiding the use of more invasive techniques.

CLINICAL PROCEDURE

OCT B-scan images from a cohort comprising Neurodegenerative Disease patients and healthy subjects were compiled for this study. The prospective study protocol and procedures were approved by the Aragon Research Ethics Committee (CEICA) according to Declaration of Helsinki. Informed consent was obtained from all participants included in the study. In all cases, participant inclusion and data collection, storage, and processing complied with the relevant data protection legislation (Spanish Law 15/1999 on the protection of personal data, Spanish Law 14/2007 on biomedical research, and European Regulation 2016/679 on the processing of personal data). Neurodegenerative disease cohorts were recruited from the Neurology Outpatient Clinic at the MSUH. PD patients had to meet the criteria established by the Parkinson’s UK Brain Bank. [(21)] The MS cohort comprised patients diagnosed with MS under standard clinical and neuroimaging criteria. [(22)] The participants in the healthy control cohort were spouses or relatives of the patients receiving routine care in the Ophthalmology Outpatient Clinic at the MSUH.

Before their inclusion in the study, all subjects were evaluated by means of clinical interviews and a complete ophthalmological examination to rule out the following systemic and ophthalmological alterations that could influence the OCT measurements (as seen in other studies): high refractive defects, opacity in ocular media, prior intraocular surgery, high intraocular pressure (IOP), or glaucoma, optic neuropathies, diabetes mellitus, or any cause of visual loss not attributable to the diseases of interest (PD or MS). [(13, 17, 18)] Retinal OCT images were recorded by an experienced technician using swept-source (SS) OCT technology and the HD-line examination protocol with the Tru-Track eye-tracking technology available with the Deep Range Imaging (DRI) Triton (Topcon Corporation, Tokyo, Japan) SS-OCT device. All these clinical procedures were developed according towere performed in accordance with guidelines and regulations.

An independent ophthalmologist (O1) participated in the first phase of the study described below. He selected and exported high-quality OCT images according to the proposed extension to the OSCAR IB and APOSTEL guidelines for OCT images and artificial intelligence (AI) and analyzed the following data: [(16)] In total, 30 OCT B-scan images taken from MS patients were selected: gender ratio 15:15 (female:male), eye ratio 14:16 (right:left), mean age 61.95 (± 10.09); 27 OCT B-scan images were taken from PD patients: gender ratio 14:13, eye ratio 13:14, mean age 70.72 (± 6.20); and 47 OCT B-scan images were taken from age-matched healthy subjects: gender ratio 23:24, eye ratio 22:25, mean age 67.57 (± 8.13). Gender, age, and intraocular pressure were considered as potential confounding factors in the study. Therefore, before statistical analyses were performed, it was verified that there were no significant differences in these variables between the study groups. All groups presented p > 0.05.

Ophthalmologist O1 exported OCT B-scan images of one eye per patient from the above-mentioned cohorts. The images were exported in TIFF format at a resolution of 1024 x 992 pixels, as shown in Fig. 1, and were randomly numbered to ensure that the other ophthalmologist (O2) was blind to the group (pathological or healthy) during monitoring of the processing result.

IMAGE-PROCESSING ALGORITHM

Superpixel segmentation imaging [(23)] was employed as the main kernel to process the OCT retinal images. Briefly, the superpixel technique groups pixels with similar structural properties to generate clusters with similar meaningful properties. A custom script was written in Matlab^™ to execute five supervised semi-automatic steps as shown in Fig. 2.

Step #1: The 24-bit RGB .tiff images (1024 x 992 pixels) exported by O1 from the OCT device were converted to 8-bit .tiff format. The converted images were cropped to form a square (992 x 992 pixels) to allow creation of a mathematical matrix during image processing if required. The raw images from the OCT device were saved in the working directory. Step #2: Gaussian low-pass filtering and the frequency–domain median filter were applied to remove speckle noise and power frequency artifacts, respectively, from the raw images converted in Step #1. Image contrast was then adjusted by histogram equalization to provide as flat an image histogram as possible. Step #3: Once contrast enhancement and noise filtering have been applied to the OCT images, the main function of the script calls the “superpixels” Matlab function where the number of superpixels, the number of iterations, and the regularity of the shape of the detected superpixels are parameters predefined by O2 and set for all the analyzed images. Next, execution of the algorithm shows the numbered computed superpixels under the predefined conditions described above. At this point, O2 supervises the region of interest (choroid tissue) to be clustered by selecting those numbered superpixels that will be grouped into a single segmented retinal region as described in Step #4. Step #4: The algorithm automatically clusters the retinal structure selected by O2 by calling the “regionpropos” Matlab function, which is corrected by supervised expert ophthalmologist to provide optimal segmentation of the choroidal layer. Step #5: Finally, the clustered region of interest (i.e. the choroid), the optical density automatically computed by the script (measured as standard mean gray level), and total area of the supervised choroidal clusters are computed.

Based on the computed outputs, three parameters were defined to quantify the structural changes to the choroid due to neurodegeneration: choroidal area (CA), defined as the total area corresponding to the segmented choroidal region of interest (pixel², px²). Choroidal optical image density (COID), defined as the mean pixel value of the segmented area. And choroidal density (CD), calculated as the ratio between CA and COID, that is the optical density of the image per unit area.

STATISTICAL ANALYSIS

Statistical analysis was performed using the IBM SPSS V.20 statistical software. Normal data distribution was detected using the Shapiro–Wilk test and parametric tests were used for data analysis. The one-way ANOVA test was used for multiple comparisons, and post-hoc analysis was used to identify statistical differences between groups. There was considered to be a statistically significant difference if the p-value was less than 0.05.

A total of 104 OCT B-scan images were processed and choroidal tissue was identified and delimited using the superpixel segmentation (SpS) algorithm. These results were analyzed according to the original group to which they belonged. Figure 3 shows the choroid boundaries defined by the algorithm in an OCT B-scan image, making the differences in optical density between the different retinal layers clearly visible.

The image obtained by algorithmic processing (Fig. 3c) shows irregularities along the lower boundary. SpS detects variations in semantic structure and delimits the area with these changes within the image. This allows definition of structure edges, in this case choroid boundaries. The processed images (Fig. 3c) show that the cross section selected to measure the thickness varies greatly depending on its position on the 10 mm horizontal axis because the choroid boundaries vary throughout the B-scan.

The following figures show the image processing output for 3 subjects chosen at random in each of the groups. The upper image (Fig. 4a) shows the algorithm’s output for a healthy subject, while the middle and lower images show the algorithm’s output for MS (Fig. 4b) and PD (Fig. 4c) patients, respectively.

While there are no significant visually perceptible differences in the B-scan image (left-hand images in Fig. 4), processing with the SpS algorithm makes the boundaries of the choroid evident. It is possible to visually detect differences between Fig. 4a (healthy eye) and the other images (pathological eyes, Fig. 4b and Fig. 4c). The algorithm parameters mentioned above also identify clear differences according to subject type in these cases: the healthy eye (Fig. 4a) presents CA = 154940 px², COID = 51.22 standard mean pixel gray level, and CD = 3025.15; the MS eye (Fig. 4b) presents CA = 112761 px², COID = 62.40 standard mean pixel gray level, and CD = 1807.18; and the PD eye (Fig. 4c) presents CA = 82072 px², COID = 70.95 standard mean pixel gray level, and CD = 1358.22.

The next image (Fig. 5) shows SpS processing on a B-scan image from a tilted retina. As can be seen, the algorithm is not affected by OCT B-scan image positioning issues due to the shape of the retina during the examination.

Algorithmic choroid analysis produced three calculated parameters. These results were analyzed according to the original group to which they belonged. The next image (Fig. 6) shows a summary of the results obtained for each group, represented by the mean and the standard deviation.

CA is greatly decreased in Neurodegenerative Disease patients versus healthy subjects (p < 0.001). COID is higher in the Neurodegenerative Disease group than in the control group (p < 0.001), and CD, which is related to choroidal area boundaries, is lower in the Neurodegenerative Disease cohort than in the control group (p < 0.001). One-way ANOVA analysis showed statistically significant differences between groups. Table 1 shows the mean and the standard deviation in each group and the post-hoc analysis differences in significance. All analyzed parameters, except COID, showed significant differences between the MS and PD groups.

Table 1

Mean ± standard deviation of the proposed parameters for each group obtained from images processing and significance of the comparison between the three groups with the Post-Hoc analysis results.
	Control Group	MS cohort	PD cohort	Significance
CA	154071.53 ± 20948.58	117736.13 ± 16153.50	94722.85 ± 19510.99	p < 0.001(C vs MS p < 0.001, C vs PD p < 0.001, MS vs PD p < 0.001)
COID	57.14 ± 6.09	64.49 ± 7.71	69.90 ± 6.74	P < 0.001 (C vs MS p < 0.001, C vs PD p < 0.001, MS vs PD p = 0.013)
CD	2784.82 ± 562.93	1858.95 ± 370.34	1354.36 ± 395.77	P < 0.001 (C vs MS p < 0.001, C vs PD p < 0.001, MS vs PD p = 0.001)

Measurements of choroid area (CA): in pixel square (px ² ). Measurement of choroid optical image density (COID): standard mean pixel grey level. Choroid density (CD) is an dimensionless parameter. Abbreviations: CA: choroid area; COID: choroid optical image density; CD: choroid density; vs: versus; MS: multiple sclerosis; PD: Parkinson disease.

In recent years, research groups have focused on developing new early diagnosis and monitoring strategies for Neurodegenerative Disease based on objective biomarkers. They have especially concentrated on alternative non-invasive techniques that are less expensive, safer, and more comfortable for patients than lumbar puncture to remove cerebrospinal fluid or MRI with intravenous contrast. OCT has demonstrated it can detect neuronal alterations in retinal tissue in MS, PD, and AD patients’ eyes. These defects are considered disease biomarkers and are directly related to the degree of disease severity, the number of years since onset, and the severity scales used by neurologists to monitor Neurodegenerative Disease. [(4, 18)] Optical image-based diagnosis techniques have evolved and test performance time and the diagnostic sensitivity of the procedures have improved. Researchers have created algorithms that facilitate clinicians’ work by processing these large masses of raw data and delivering diagnoses. [(9, 24, 25)] Artificial intelligence and the associated algorithms are capable of detecting the minuscule alterations that make it possible to classify a subject as healthy or pathological. The advances in diagnostic processing techniques mean only raw image information is required, making the procedure independent of image quality and uninfluenced by external suggestions. Thus, although poor image quality affects examiners’ decisions, the algorithm’s decisions remain constant. [(26)]

This paper strives to advance the research group's discoveries and developments in relation to retinal biomarkers of Neurodegenerative Disease, implementing these new image-processing techniques independent of the examiner and advancing the body of knowledge on diagnosis of these diseases. Recent studies have sought to identify possible vascular changes in Neurodegenerative Disease. A new algorithm based on superpixel segmentation image processing has been developed to study retinal vascular structures from OCT B-scan images regardless of the instrument employed. This algorithm can be used with any B-scan image generated by any commercial OCT device. In this study, only OCT B-scan images obtained with a Triton® SS-OCT were used. [(4, 27, 28)] A custom algorithm was then applied to detect retinal vascular tissue, specifically choroidal tissue, and identify possible differences between healthy subjects and Neurodegenerative Disease patients. Cohort members were conscientiously selected to avoid bias caused by potential confounding factors: gender, age, eye ratio, and IOP.^46–50 Furthermore, only B-scan examinations performed in the morning were included to avoid possible changes associated with the circadian rhythm. [(29)]

The results show that the processing algorithm is able to detect tissue changes in the image propierties and can delimit the choroidal area in all subject types. Analyzing the processing images shows how those zones differ between cohorts. Based on the image data, the algorithm calculates three parameters according to the choroid image characteristics, presenting significant differences between study groups. The SpS algorithm shows a good capacity to differentiate between Triton OCT images taken from Neurodegenerative Disease and healthy subjects according to a neurologist’s diagnosis, and also to differentiate between PD and MS eyes. This provides the potential for differential diagnosis between different Neurodegenerative Diseases.

Prior studies on choroidal tissue alterations in Neurodegenerative Disease patients show similar results. Uppugunduri et al previously published an automated technique using the open-source ImageJ software to process OCT B-scan images to detect the Haller and Sattler layers in choroidal tissue. However, most published research uses the software provided with the different commercial OCT devices to study the choroid. As in previous studies, this paper's results show a smaller area of choroidal tissue than in healthy subjects. It has also been possible to detect variation in optical density in MS patient choroids versus healthy eyes. Esen et al (2016) and Garcia et al (2018) analyzed choroidal thickness parameters obtained with different commercial OCT software from MS patients versus healthy subjects and detected that MS eyes presented a significant decrease versus healthy eyes. [(13, 14, 15)] Different hypotheses suggest that changes can be found in the vascular structure in MS: ganglion cell death in MS is reflected in a decrease in tissue volume and in metabolic demand to carry out its functions, which can translate as a decrease in blood flow and, therefore, in volume. Also, due to the inflammatory process associated with the disease, the walls of the vessels are damaged, which can manifest as a loss of tissue. [(30, 31, 32)] The results in this study show a reduction in the optical density of images from MS patients that could be related to these physiological changes in blood flow and vessel properties in MS. It would be beneficial to have histological studies of healthy subjects’ and Neurodegenerative Disease patients’ eyes to test these hypotheses and to better understand the pathophysiological mechanisms affecting tissues, in particular vascular tissues such as the choroid.

The results show a smaller area of choroidal tissue in PD patients than in healthy subjects and MS patients. Moreover, optical density in PD patient choroids varies versus healthy or MS eyes. Choroidal tissue study in PD patients is a subject that generates a lot of controversy because the findings differ widely between authors. Garcia et al (2017) detected choroid thickening in the peripapillary zone in PD and Satue et al (2017) made the same findings in macular zone thickness. However, when using angiography mediated by OCT technology they did not find any differences in vasculature dynamics and structure between groups. [(13, 17, 33)] This could be because the tissue structures observed by OCT in the choroid differ from the analysis of purely vascular structures and, mainly, of flux density in the deep and superficial choroidal plexus. Conversely, Eraslan et al (2016) and Moschos et al (2017) found a decrease in choroidal thickness in PD patients' eyes in results obtained using SD-OCT outputs. [(19, 20)]

Previous studies used the automatic or semiautomatic OCT segmentation technique that delimits choroidal thickness between the BM and the sclerochoroidal interface These variations in findings in the studies may be because OCT measures thicknesses and distances at different points and, as can be seen in the OCT B-scan images obtained in this study, the boundary of the choroid plexus does not have a regular edge. OCT requires performance of an interference measurement at each point that it wants to calculate. Moreover, the process has to be repeated in each A-scan along the B-scan length. Some OCT systems are able to perform up to 27000 A-scan per sec in the best exam conditions, when the patient's eye does not present any movement and he is able to collaborate optimally throughout the procedure. In this case, if these A-scan measurements were totally real, it would take several minutes to perform each eye examination. OCT acquisition, however, takes no more than 3 seconds. Using a custom-built OCT system in laboratory research makes it possible to dedicate the necessary time to obtaining a precise measurement at each point (A-scan) because the analyzed sample does not move or get tired; in vitro models, not patients, were used in the laboratory. In relation to this, consideration should be given to whether the data-processing function of the OCT software could calculate the weighting of the values based on the data measured. We do not know what the limit of the choroid is and its detection by OCT, as the results of this study show (Fig. 4), the choroidal boundary image properties is irregular and, in these cases, the thickness measurement could be influenced by the points where the measurement is taken and by the data extrapolation areas. A specific data such as thickness in tissues that are difficult to access may not be the most appropriate to characterize a structure.

Kwapong et al (2018) analyzed microvascular dynamics through angiography mediated by OCT techniques and reported zones with decreased microvascular density in PD patients versus healthy subjects. [(34)] More recently, Robbins et al (2021) published their results about PD choroidal dynamics and structure based on angiography mediated by OCT images. They processed these images using the open source ImageJ software and performed choroidal tissue measurements using software tools. They defined two parameters of interest: subfoveal choroidal thickness (SFCT), which is the linear distance between the outer border of the retinal pigment epithelium perpendicular to the hyper-reflective sclerochoroidal junction manually drawn on the image; and choroidal area (CVI), which was calculated by dividing the luminal area by the total choroidal area, both of which were manually drawn on the image. Their findings showed a reduction in both parameters in PD versus healthy eyes. [(35)] More recently, Zhang et al (2022) used the Uppugunduri AI algorithm to analyze OCT and OCT-mediated angiography images taken from PD and healthy subjects’ eyes. Again, their study of choroidal dynamics, blood flow, and vessel volume showed significant differences between the two groups. Also, choroidal thickness was significantly decreased in PD eyes. [(36)]

The results obtained in this study using an independent algorithm show changes in choroidal tissue in Neurodegenerative Disease patients versus healthy eyes. Firstly, it is evident that the boundaries defining the choroid delimit a smaller area (Fig. 4). The calculated CA parameter is significantly lower in PD patients than in MS patients, and in MS patients than in healthy eyes. Secondly, as reported by Zhang et al (2022), optical density analysis of the image (COID) shows significant differences in the choroidal vascular structure in Neurodegenerative Disease patients versus healthy eyes, and the relationship between the COID and CA, choroidal density, is also significantly lower in PD patients than in MS patients, and in MS patients than in healthy eyes. It is important to highlight the information extracted from the properties of the image, indirectly from the studied tissue, the change that we observe between healthy and pathological eyes, between MS and PD eyes.

Pathophysiological choroidal changes in PD are believed to be associated with dopamine levels, which could affect blood perfusion. [(37, 38)] Also, PD patients have increased cerebral small vessels, and decreased retinal dimensions have been observed in these conditions. [(39)] Moreover, α-syn-GFP deposition around retinal arterial vessels could be related to changes in the capillary plexus. [(34, 40)]

OCT devices detect interferometry between reference and measurement beams and, based on this information, calculate the distance travelled. In optics, distance is not only a geometrical measurement but an optical distance that depends on index refraction and geometrical distance. If the media through which light propagates change, events occur that change the optical distance. Various studies put forward different hypotheses regarding changes in vascular tissue, blood perfusion, vessel walls, etc. in PD patients. A significant change in refractive index is necessary to produce a detectable change in light propagation; however, it cannot be ruled out that these variations exist and that they produce variations in the optical distance calculated by OCT in the interferometry analysis. An example would be a swollen cornea and how it alters results obtained using optical diagnosis techniques. Also, commercial OCT devices do not usually consider changes in the direction of propagation in the path of the examination beam due to changes in the medium within the eye. This fact could also produce differences in the optical distances calculated, and therefore, in the thickness measurements.

For this reason, we propose not focusing on an isolated measurement of uncertain origin, such as the thickness of the choroidal structure. It is evident that the extent of the discrepancies and variability in the results is due to reasons as yet unknown. The area parameter, the optical density image, and other parameters previously mentioned and obtained automatically from an OCT image are more reliable because they allow elimination of many of the factors listed above. It is becoming clear that changes occur in the choroidal vascular structure in PD subjects and more studies are necessary. AI procedures can help to find these differences and define a pattern, but it is necessary to have a large number of standardized, high-quality images that can be processed. Other imaging techniques, such as MRI, have public databases containing thousands of images and the corresponding diagnoses with which to develop this type of diagnostic support algorithm based on deep learning.

The non-commercial superpixel segmentation algorithm used in this study allows detection of choroidal alterations in Neurodegenerative Disease by processing OCT images. This method may be used in early diagnosis of MS or PD and has the potential to be a non-invasive, easily tolerated, low-cost and effective tool for early diagnosis and even for differential diagnosis of Neurodegenerative Disease. In addition, it is an algorithm that can process images obtained with any OCT device, making it easy to integrate into clinical practice.

Funding: This study was supported by Carlos III Health Institute grants PI17/01726 and PI20/00437, and by the Inflammatory Disease Network (RICORS) (RD21/0002/0050) (Carlos III Health Institute).

Acknowledgements

The authors would like to acknowledge the use of the infrastructure at the Department of Ophthalmology at Miguel Servet University Hospital. We would also like to thank Juan Manuel Bueno (Optics and Nanophysics Research Center, University of Murcia) who contributed to the development and improvement of the Algorithm.

Author Contributions: "Conceptualization, S.O. and F.A.; Methodology, V.M and S.O; Software, F.A.; Validation, S.O, and E. G-M.; Formal Analysis, S.O, and E. G-M.; Investigation, S.O, and E. G-M.; Writing – Original Draft Preparation, S.O.; Writing – Review & Editing, E.G-M.; Supervision, F.A.; Project Administration, E. G-M.; Funding Acquisition, E.G-M.”

Data availability: The datasets are available from the corresponding author upon reasonable request.

Conflict of interest: No conflicting relationship exists for any author.

Klein R, Lee KE, Knudtson MD, Gangnon RE, Klein BE. Changes in visual impairment prevalence by period of diagnosis of diabetes: the Wisconsin Epidemiologic Study of Diabetic Retinopathy. Ophthalmology. 2009;116(10):1937–42.
Wong TY, Shankar A, Klein R, Klein BE, Hubbard LD. Retinal arteriolar narrowing, hypertension, and subsequent risk of diabetes mellitus. Arch Intern Med. 2005;165(9):1060–5.
Ahmadi H, Kolko M, Hamann S. Enhanced depth imaging optical coherence tomography of the optic nerve head improves correct diagnosis in glaucoma suspects without glaucomatous optic disc morphology. BMJ Case Rep. 2022;15(2).
Garcia-Martin E, Ara JR, Martin J, Almarcegui C, Dolz I, Vilades E, et al. Retinal and Optic Nerve Degeneration in Patients with Multiple Sclerosis Followed up for 5 Years. Ophthalmology. 2017;124(5):688–96.
Gordon-Lipkin E, Chodkowski B, Reich DS, Smith SA, Pulicken M, Balcer LJ, et al. Retinal nerve fiber layer is associated with brain atrophy in multiple sclerosis. Neurology. 2007;69(16):1603–9.
Archibald NK, Clarke MP, Mosimann UP, Burn DJ. Retinal thickness in Parkinson's disease. Parkinsonism Relat Disord. 2011;17(6):431–6.
Marziani E, Pomati S, Ramolfo P, Cigada M, Giani A, Mariani C, et al. Evaluation of retinal nerve fiber layer and ganglion cell layer thickness in Alzheimer's disease using spectral-domain optical coherence tomography. Invest Ophthalmol Vis Sci. 2013;54(9):5953–8.
Lynch M, Pham W, Sinclair B, O'Brien TJ, Law M, Vivash L. Perivascular spaces as a potential biomarker of Alzheimer's disease. Front Neurosci. 2022;16:1021131.
Montolío A, Cegoñino J, Garcia-Martin E, Pérez Del Palomar A. Comparison of Machine Learning Methods Using Spectralis OCT for Diagnosis and Disability Progression Prognosis in Multiple Sclerosis. Ann Biomed Eng. 2022;50(5):507–28.
Koronyo-Hamaoui M, Koronyo Y, Ljubimov AV, Miller CA, Ko MK, Black KL, et al. Identification of amyloid plaques in retinas from Alzheimer's patients and noninvasive in vivo optical imaging of retinal plaques in a mouse model. Neuroimage. 2011;54 Suppl 1:S204-17.
Chalkias IN, Tegos T, Topouzis F, Tsolaki M. Ocular biomarkers and their role in the early diagnosis of neurocognitive disorders. Eur J Ophthalmol. 2021;31(6):2808–17.
Uppugunduri SR, Rasheed MA, Richhariya A, Jana S, Chhablani J, Vupparaboina KK. Automated quantification of Haller's layer in choroid using swept-source optical coherence tomography. PLoS One. 2018;13(3):e0193324.
Garcia-Martin E, Jarauta L, Pablo LE, Bambo MP, Ara JR, Martin J, et al. Changes in peripapillary choroidal thickness in patients with multiple sclerosis. Acta Ophthalmol. 2019;97(1):e77-e83.
Garcia-Martin E, Jarauta L, Vilades E, Ara JR, Martin J, Polo V, et al. Ability of Swept-Source Optical Coherence Tomography to Detect Retinal and Choroidal Changes in Patients with Multiple Sclerosis. J Ophthalmol. 2018;2018:7361212.
Esen E, Sizmaz S, Demir T, Demirkiran M, Unal I, Demircan N. Evaluation of Choroidal Vascular Changes in Patients with Multiple Sclerosis Using Enhanced Depth Imaging Optical Coherence Tomography. Ophthalmologica. 2016;235(2):65–71.
Petzold A, Albrecht P, Balcer L, Bekkers E, Brandt AU, Calabresi PA, et al. Artificial intelligence extension of the OSCAR-IB criteria. Ann Clin Transl Neurol. 2021;8(7):1528–42.
Satue M, Castro L, Vilades E, Cordon B, Errea JM, Pueyo A, et al. Ability of Swept-source OCT and OCT-angiography to detect neuroretinal and vasculature changes in patients with Parkinson disease and essential tremor. Eye (Lond). 2022.
Satue M, Rodrigo MJ, Obis J, Vilades E, Gracia H, Otin S, et al. Evaluation of Progressive Visual Dysfunction and Retinal Degeneration in Patients With Parkinson's Disease. Invest Ophthalmol Vis Sci. 2017;58(2):1151–7.
Eraslan M, Cerman E, Yildiz Balci S, Celiker H, Sahin O, Temel A, et al. The choroid and lamina cribrosa is affected in patients with Parkinson's disease: enhanced depth imaging optical coherence tomography study. Acta Ophthalmol. 2016;94(1):e68-75.
Moschos MM, Chatziralli IP. Evaluation of Choroidal and Retinal Thickness Changes in Parkinson's Disease Using Spectral Domain Optical Coherence Tomography. Semin Ophthalmol. 2018;33(4):494–7.
Litvan I, Goldman JG, Tröster AI, Schmand BA, Weintraub D, Petersen RC, et al. Diagnostic criteria for mild cognitive impairment in Parkinson's disease: Movement Disorder Society Task Force guidelines. Mov Disord. 2012;27(3):349–56.
Polman CH, Reingold SC, Banwell B, Clanet M, Cohen JA, Filippi M, et al. Diagnostic criteria for multiple sclerosis: 2010 revisions to the McDonald criteria. Ann Neurol. 2011;69(2):292–302.
Ren X, Malik, J. Learning a classification model for segmentation. In: ICCV. pp 10–17. IEEE Computer Society 20032003. p. pp. 10–7
Belić M, Bobić V, Badža M, Šolaja N, Đurić-Jovičić M, Kostić VS. Artificial intelligence for assisting diagnostics and assessment of Parkinson's disease-A review. Clin Neurol Neurosurg. 2019;184:105442.
Warren SL, Moustafa AA. Functional magnetic resonance imaging, deep learning, and Alzheimer's disease: A systematic review. J Neuroimaging. 2022.
Stanciu SG, Ávila FJ, Hristu R, Bueno JM. A Study on Image Quality in Polarization-Resolved Second Harmonic Generation Microscopy. Sci Rep. 2017;7(1):15476.
Garcia-Martin E, Bambo MP, Marques ML, Satue M, Otin S, Larrosa JM, et al. Ganglion cell layer measurements correlate with disease severity in patients with Alzheimer's disease. Acta Ophthalmol. 2016;94(6):e454-9.
Polo V, Satue M, Rodrigo MJ, Otin S, Alarcia R, Bambo MP, et al. Visual dysfunction and its correlation with retinal changes in patients with Parkinson's disease: an observational cross-sectional study. BMJ Open. 2016;6(5):e009658.
Usui S, Ikuno Y, Akiba M, Maruko I, Sekiryu T, Nishida K, et al. Circadian changes in subfoveal choroidal thickness and the relationship with circulatory factors in healthy subjects. Invest Ophthalmol Vis Sci. 2012;53(4):2300–7.
Wang X, Jia Y, Spain R, Potsaid B, Liu JJ, Baumann B, et al. Optical coherence tomography angiography of optic nerve head and parafovea in multiple sclerosis. Br J Ophthalmol. 2014;98(10):1368–73.
Plumb J, McQuaid S, Mirakhur M, Kirk J. Abnormal endothelial tight junctions in active lesions and normal-appearing white matter in multiple sclerosis. Brain Pathol. 2002;12(2):154–69.
Doche E, Lecocq A, Maarouf A, Duhamel G, Soulier E, Confort-Gouny S, et al. Hypoperfusion of the thalamus is associated with disability in relapsing remitting multiple sclerosis. J Neuroradiol. 2017;44(2):158–64.
Satue M, Obis J, Alarcia R, Orduna E, Rodrigo MJ, Vilades E, et al. Retinal and Choroidal Changes in Patients with Parkinson's Disease Detected by Swept-Source Optical Coherence Tomography. Curr Eye Res. 2018;43(1):109–15.
Kwapong WR, Ye H, Peng C, Zhuang X, Wang J, Shen M, et al. Retinal Microvascular Impairment in the Early Stages of Parkinson's Disease. Invest Ophthalmol Vis Sci. 2018;59(10):4115–22.
Robbins CB, Grewal DS, Thompson AC, Powers JH, Soundararajan S, Koo HY, et al. Choroidal Structural Analysis in Alzheimer Disease, Mild Cognitive Impairment, and Cognitively Healthy Controls. Am J Ophthalmol. 2021;223:359–67.
Zhang Y, Yang L, Gao Y, Zhang D, Tao Y, Xu H, et al. Choroid and choriocapillaris changes in early-stage Parkinson's disease: a swept-source optical coherence tomography angiography-based cross-sectional study. Alzheimers Res Ther. 2022;14(1):116.
Nickla DL, Totonelly K, Dhillon B. Dopaminergic agonists that result in ocular growth inhibition also elicit transient increases in choroidal thickness in chicks. Exp Eye Res. 2010;91(5):715–20.
Huemer KH, Zawinka C, Garhöfer G, Golestani E, Litschauer B, Dorner GT, et al. Effects of dopamine on retinal and choroidal blood flow parameters in humans. Br J Ophthalmol. 2007;91(9):1194–8.
van der Holst HM, van Uden IW, Tuladhar AM, de Laat KF, van Norden AG, Norris DG, et al. Cerebral small vessel disease and incident parkinsonism: The RUN DMC study. Neurology. 2015;85(18):1569–77.
Price DL, Rockenstein E, Mante M, Adame A, Overk C, Spencer B, et al. Longitudinal live imaging of retinal α-synuclein::GFP deposits in a transgenic mouse model of Parkinson's Disease/Dementia with Lewy Bodies. Sci Rep. 2016;6:29523.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Detecting Structural changes in the retina of neurodegenerative patients through OCT supervised segmentation

Status:

Version 1

Abstract

Background

Methods

Results

Conclusions

Figures

Introduction

Material And Methods

CLINICAL PROCEDURE

IMAGE-PROCESSING ALGORITHM

STATISTICAL ANALYSIS

Results

Discussion

Declarations

References

Additional Declarations

Status:

Version 1