Automated anterior chamber inflammation assessment using numerical computation program and image processing algorithm

Abstract

The invention relates to a computer-implemented method in which anterior chamber inflammation is automatically assessed using a numerical computation program and an image processing algorithm.

Description

[0001] DESCRIPTION

[0002] AUTOMATED ANTERIOR CHAMBER INFLAMMATION

[0003] ASSESSMENT USING NUMERICAL COMPUTATION PROGRAM AND

[0004] IMAGE PROCESSING ALGORITHM

[0005] Technical Field

[0006] The invention relates to a computer-implemented method in which anterior chamber inflammation is automatically assessed using a numerical computation program and an image processing algorithm.

[0007] Prior Art

[0008] While the biomicro scopic physical examination method is used in routine medical practice for the assessment of anterior chamber inflammation, other methods in the literature and some academic studies include laser flare and cell photometry (LFP), anterior segment fluorophotometry, ultrasonic biomicroscopy (UBM), bioluminescence, and anterior segment optical coherence tomography (OCT) systems.

[0009] In addition, manual methods or automated algorithms with Image J1,2, Python3, 4, and ImagelQ5have been used to assess anterior chamber inflammation.

[0010] The techniques used have problems such as:

[0011] • being operator-dependent and having low inter-operator agreement,

[0012] • not being supported by objective data but based on clinical experience and subjectivity,

[0013] • not allowing grading of the patient’s clinical condition in terms of progression, regression, treatment response, and medico-legal medical archiving,

[0014] • not performing weighting according to anterior chamber volume, • difficulty in assessment in cases where the optical medium is blurred, such as corneal opacities and anterior chamber reaction,

[0015] • the average depth of the anterior chamber being 3 mm, but the assessment being carried out over 1x1 mm sections while disregarding the rest of the anterior chamber,

[0016] • inability to subtype the cells,

[0017] • inability to perform flare assessment, and the requirement of an LFP device for this process, while in routine medical practice LFP analyses are timeconsuming and have low repeatability.

[0018] CN116965768A discloses a system that automatically and quantitatively analyzes the degree of intraocular anterior chamber inflammation.

[0019] RU2754292C1 discloses a method that predicts the long-term high risk of autoimmune inflammation of post-traumatic uveitis disease.

[0020] In the document entitled Automated Quantitative Analysis of Anterior Segment Inflammation Using Swept-Source Anterior Segment Optical Coherence Tomography: A Pilot Study, the automatic quantitative analysis of anterior segment inflammation using Swept-Source anterior segment optical coherence tomography is described.

[0021] When the studies in the state of the art are examined, a need has arisen for the development of a computer-implemented method in which anterior chamber inflammation is automatically assessed using a numerical computation program and an image processing algorithm.

[0022] Objectives of the Invention

[0023] The object of the present invention is to develop a computer-implemented method in which anterior chamber inflammation is automatically assessed using a numerical computation program and an image processing algorithm. Another object of the present invention is to develop a computer-implemented method that uses optical coherence tomography images as objective data and allows quantitative calculations.

[0024] Another object of the present invention is to develop a computer-implemented method that enables rapid assessment of multiple sections of a patient using batch processing and, when the number of sections is increased, allows not only sectional but theoretically volumetric analysis.

[0025] Detailed Description of the Invention

[0026] The invention relates to a computer-implemented method in which anterior chamber inflammation is automatically assessed using a numerical computation program and an image processing algorithm, it comprises; converting images obtained from patients using an optical coherence tomography (OCT) device into grayscale format, performing corneal segmentation on the grayscale-converted images using a Non-local Means (NLM) noise reduction filter, detecting object edges of the images after segmentation using the Chan- Vese method with the active contour function and obtaining a binarized image, selecting regions of interest (ROIs) corresponding to air and the anterior chamber on the binarized image using the bwpropfilt function, determining the region of interest (ROI) corresponding to the anterior chamber and applying a median filter to this area, calculating the mean brightness reflectance of the region corresponding to the anterior chamber, selecting the region of interest (ROI) corresponding to air and applying a low-pass filter, calculating the mean brightness reflectance of the region representing air, calculating the aqueous-air relative density index based on the anterior chamber and air reflectance.

[0027] The first stage of the method of the present invention is loading the patients’ OCT images into the numerical computation program interface. Successful segmentation of the cornea in the loaded images is of great importance for the subsequent algorithmic calculations. The numerical computation program provides a wide range of functions and filters in its image processing package. In the method of the present invention, it has been determined that the most accurate corneal segmentation can be performed using the "Non-local Means Filter" (NLMF) deionization filter and the localized segmentation of "activecontour." The NLMF deionization filter estimates its own degree of smoothing based on the standard deviation of noise in the image. It removes noise while preserving the sharpness of strong edges, such as those of the cornea. Due to this feature, it is also used in images in the field of astronomy. The "activecontour" segmentation is a method that aims to detect the edges, shape, and stereo features of an object in 2D images in the best possible way through repeated morphological operations. It is currently used effectively in real-time imaging technologies, such as traffic sign recognition and tracking approaching or receding vehicles in flowing traffic. The contrast of the resulting binarized image is increased to high levels, and small objects in the image are filtered out. In this way, a binarized image is obtained that overlaps with the actual corneal image and can later be used as a mask.

[0028] After successful segmentation of the cornea, the hyporeflective black regions corresponding to air and the anterior chamber in the binarized image are selected separately. The areas occupied by these images are defined individually as air and anterior chamber masks, and the algorithm continues to operate within the regions of interest (ROIs) determined by these masks.

[0029] The ROI mask corresponding to the anterior chamber is selected, and the ratio of this mask to the entire image is defined as the anterior chamber volume in percentage terms. A low-pass filter is applied within this area, and objects small enough to pass through this filter are defined as cells. These cells are counted, and their morphological features are calculated. Dividing the number of cells by the anterior chamber volume is defined as the cell density. The area and eccentricity of the cells are calculated individually and as average values within the program. The mean brightness reflectance of the region representing the anterior chamber is also calculated.

[0030] The ROI mask corresponding to air is selected, and the mean brightness reflectance of the air is calculated. The aqueous-air relative density index is then calculated based on the anterior chamber and air reflectance. While this index was previously calculated in earlier techniques semi-automatically using manually selected rectangular areas in the air and anterior chamber, in the method of the present invention it is fully automated.

[0031] The aqueous-air relative density index is an intraocular inflammation marker that can be subjectively assessed and graded in clinical practice under biomicro scopic examination, and objectively evaluated using laser flare cell photometry (LFP). These methods rely on assessing the scattered light, as proteins that leak into the aqueous humor during inflammation scatter the biomicroscope or LFP light beams.

[0032] The proteins that leak are considered a marker of the disrupted intraocular blood- aqueous barrier resulting from inflammation.

[0033] The numerical computation program is a programming and numerical calculation platform with wide applications in the engineering world, and image processing is one of these applications. In this study, an algorithm was developed in the numerical computation program by using the applications and functions of the image processing package to perform operations and calculations on images obtained from the OCT device. The algorithm can be run on a single image input or used in Batch Processing to analyze a large number of input images in series. When executed, the algorithm calculates the number, size, and eccentricity of cells in the anterior chamber, the percentage area of the anterior chamber relative to the entire image, and the aqueous-air relative density index, and it generates the results in the application workspace.

[0034] The implementation stages of the method of the present invention are explained in more detail below.

[0035] Preprocessing and Corneal Binarization

[0036] Image processing generally requires various preprocessing steps such as histogram adjustments, filters, and binarization methods, and is used as a step to prepare the image for subsequent operations and to enable functions to work more efficiently. These steps either filter out noise and artifacts or highlight specific parts of the image, ultimately creating a coarse model. This coarse model is a matrix consisting of ones and zeros, where pixel values of 1 represent white and are referred to as the foreground, and pixel values of 0 represent black and are referred to as the background.

[0037] In a 6 mm scan-length image, the cornea extends from the left edge to the right edge. With proper segmentation of the cornea, the image can be divided into three different regions of interest: hyperreflective cornea, hyporeflective air, and anterior chamber. By standard, digital images are composed of various pixels, each represented by a vector of three numbers corresponding to the three primary color channels (red, green, and blue). First, the method converts the original images into grayscale format, thereby reshaping the pixels according to light intensity values ranging from 0 to 255. Next, contrast is adjusted to saturate the top 1% and bottom 1% of image pixels to reduce noise, and a specialized denoising filter called NLMF is applied. Thereafter, the image is binarized to create a foundation for subsequent processing steps.

[0038] Dividing the image into three different regions of interest (ROIs) Since there are numerous individual hyper-reflective points scattered throughout the corneal stroma that are not connected, at this stage — or if the cornea is inadequately masked — the progression of the algorithm can result in significant errors. The initially obtained corneal binarization requires further processing to align with the true corneal contours by correcting sharp boundaries in the corneal stroma and closing gaps by merging them.

[0039] For this purpose, the binarized corneal mask is first expanded using morphological operations (‘hole’, ‘thicken’, ‘bridge’) and then reduced using the "Chan-Vese" method with the activecontour function until it correctly overlaps with the corneal boundaries and adjacent objects. This method sets the cornea as the foreground by merging it with objects located very close to the cornea and any contacting artifacts.

[0040] This method also accounted for the interference from reflection artifacts to some extent.The hyporeflective spaces in front of the cornea represent air. To select these regions, the three largest ROIs on the image are chosen using bwpropfilt (two corresponding to air spaces and one to the anterior chamber).

[0041] The bwpropfilt function allows extraction of foreground objects in binarized images based on various properties. In the method of the present invention, the topographic features present in an ideal OCT cross-section were utilized. In an ideal anterior segment OCT slice, one or two air pockets are observed in the upper right and / or left corners in front of the convex-shaped cornea, and one anterior chamber space is observed behind the cornea. Using this function, the three objects with the largest perimeters in the image are identified, and from these, the objects located in the upper right and / or left corners determined based on the image coordinate system are selected and defined as air. After removing the air and cornea from the image, the remaining region is defined as the anterior chamber.

[0042] When the images are examined, air is always present in the upper right and left corners. Selecting two ROIs that include these comers provides an effective means of masking the air. In the grayscale image, the reflectance of air is expressed as ARIi and is calculated as follows:

[0043] ARIi = mean2(brightness reflectance of the air mask) x 1020 x 640 (total image resolution) nnz (total number of pixels in the air mask) (Formula I)

[0044] From this point onward, the image can be analyzed in three regions: a single foreground object defined as the cornea, the hyporeflective area in front of the cornea representing air, and the hyporeflective area behind the cornea representing the anterior chamber.

[0045] In the grayscale image, the reflectance of the anterior chamber is expressed as ARE and is calculated as follows:

[0046] ARI2 = mean2(brightness reflectance of the anterior chamber mask) x 1020 x 640 (total image resolution) nnz (total number of pixels in the anterior chamber mask) (Formula II)

[0047] The aqueous-air relative density (ARI) index is the ratio of ARI2 to ARIi multiplied by 100.

[0048] Cell counting, density, and morphological features

[0049] After the mask corresponding to the anterior chamber is defined as the ROI, a 2x2 pixel median filter is applied to suppress noise without excessively diminishing the cells in the anterior chamber. The image is then binarized. Any object larger than 100 pixels is excluded by applying a 100-pixel low-pass filter, as such objects cannot be cells. Objects larger than I but smaller than 100 pixels are defined as cells. The cell count is defined as the number of objects remaining after applying the low- pass filter. Additionally, the eccentricity and area of each cell, along with their totals and averages, are calculated in array format. Eccentricity is the ratio of the distance between the foci of an ellipse to the length of its major axis and ranges from 0 to 1. Finally, the cells are sequentially color-coded using a red-yellow palette, enlarged to make them more distinguishable on the main image, and displayed side by side with the original image for error checking. Cell density in the images is defined as the number of cells divided by the number of pixels in the anterior chamber mask.

[0050] The method of the present invention, with an OCT cross-section of 1.96 x 6 mm, allows evaluation of a larger area in the anterior chamber than the standard I x I mm. Theoretically, if the method is integrated into OCT systems capable of imaging the entire anterior chamber, analysis of the entire anterior chamber becomes possible. In addition, it provides the ability to track progression, regression, treatment response, and enables medical archiving for medico-legal purposes.

[0051] For improved calculations, adjustments should be based on anterior chamber volume.Thanks to image processing technologies, it will be possible to define new variables that could serve as markers of anterior chamber inflammation and to examine morphological features that allow differentiation of cell subtypes.

[0052] References:

[0053] [1] Lu M, Wang X, Lei L, Deng Y, Yang T, Dai Y, et al. Quantitative Analysis of Anterior Chamber Inflammation Using the Novel CASIA2 Optical Coherence Tomography. Am J Ophthalmol. 2020; 216: 59-68. Available at: https: / / doi.Org / 10.l 016 / j.ajo.2020.03.032.

[0054] [2] Baghdasaryan E, Tepelus TC, Marion KM, Huang J, Huang P, Sadda SVR, et al. Analysis of ocular inflammation in anterior chamber-involving uveitis using swept-source anterior segment OCT. Int Ophthalmol. 2019; 39(8): 1793-180 1.

[0055] [3] Kang TS, Lee Y, Lee S, Kim K, Lee W sub, Lee W, et al. Development offully automated anterior chamber cell analysis based on image software. Sci Rep. 2021; 11(1).

[0056] [4] Keino H, Aman T, Furuya R, Nakayama M, Okada AA, Sunayama W, et al. Automated Quantitative Analysis of Anterior Segment Inflammation Using Swept- Source Anterior Segment Optical Coherence Tomography: A Pilot Study.

[0057] Diagnostics. 2022; 12(11).

[0058] [5] Sharma S, Lowder CY, Vasanji A, Baynes K, Kaiser PK, Srivastava SK. Automated Analysis of Anterior Chamber Inflammation by Spectral-Domain Optical Coherence Tomography. Ophthalmology. 2015; 122(7): 1464-1470. Available at: http: / / dx.doi.Org / 10.1016 / j.ophtha.2015.02.032.

Claims

CLAIMS1. A computer-implemented method in which anterior chamber inflammation is automatically assessed using a numerical computation program and an image processing algorithm characterized in that it comprises; converting images obtained from patients using an optical coherence tomography (OCT) device into grayscale format, performing corneal segmentation on the grayscale-converted images using a Non-local Means (NLM) noise reduction filter, detecting object edges of the images after segmentation using the Chan-Vese method with the active contour function and obtaining a binarized image, selecting regions of interest (ROIs) corresponding to air and the anterior chamber on the binarized image using the bwpropfilt function, determining the region of interest (ROI) corresponding to the anterior chamber and applying a median filter to this area, calculating the mean brightness reflectance of the region corresponding to the anterior chamber, selecting the region of interest (ROI) corresponding to air and applying a low-pass filter, calculating the mean brightness reflectance of the region representing air, calculating the aqueous-air relative density index based on the anterior chamber and air reflectance.