Convolution Neural Networks and Support Vector Machines for Automatic Segmentation of Intracoronary Optical Coherence Tomography

Cardiovascular diseases are closely associated with deteriorating atherosclerotic plaques. Optical coherence tomography (OCT) is a recently developed intravascular imaging technique with high resolution approximately 10 microns and could provide accurate quantification of coronary plaque morphology. However, tissue segmentation of OCT images in clinic is still mainly performed manually by physicians which is time consuming and subjective. To overcome these limitations, two automatic segmentation methods for intracoronary OCT image based on support vector machine (SVM) and convolutional neural network (CNN) were performed to identify the plaque region and characterize plaque components. In vivo IVUS and OCT coronary plaque data from 5 patients were acquired at Emory University with patient’s consent obtained. Seventy-seven matched IVUS and OCT slices with good image quality and lipid cores were selected for this study. Manual OCT segmentation was performed by experts using virtual histology IVUS as guidance, and used as gold standard in the automatic segmentations. The overall classification accuracy based on CNN method achieved 95.8%, and the accuracy based on SVM was 71.9%. The CNN-based segmentation method can better characterize plaque compositions on OCT images and greatly reduce the time spent by doctors in segmenting and identifying plaques.     

Repeatability of Foveal Measurements Using Spectralis Optical Coherence Tomography Segmentation Software

Purpose

To investigate repeatability and reproducibility of thickness of eight individual retinal layers at axial and lateral foveal locations, as well as foveal width, measured from Spectralis spectral domain optical coherence tomography (SD-OCT) scans using newly available retinal layer segmentation software.

Methods

High-resolution SD-OCT scans were acquired for 40 eyes of 40 young healthy volunteers. Two scans were obtained in a single visit for each participant. Using new Spectralis segmentation software, two investigators independently obtained thickness of each of eight individual retinal layers at 0°, 2° and 5° eccentricities nasal and temporal to foveal centre, as well as foveal width measurements. Bland-Altman Coefficient of Repeatability (CoR) was calculated for inter-investigator and inter-scan agreement of all retinal measurements. Spearman''s ρ indicated correlation of manually located central retinal thickness (RT₀) with automated minimum foveal thickness (MFT) measurements. In addition, we investigated nasal-temporal symmetry of individual retinal layer thickness within the foveal pit.

Results

Inter-scan CoR values ranged from 3.1μm for axial retinal nerve fibre layer thickness to 15.0μm for the ganglion cell layer at 5° eccentricity. Mean foveal width was 2550μm ± 322μm with a CoR of 13μm for inter-investigator and 40μm for inter-scan agreement. Correlation of RT₀ and MFT was very good (ρ = 0.97, P < 0.0005). There were no significant differences in thickness of any individual retinal layers at 2° nasal compared to temporal to fovea (P > 0.05); however this symmetry could not be found at 5° eccentricity.

Conclusions

We demonstrate excellent repeatability and reproducibility of each of eight individual retinal layer thickness measurements within the fovea as well as foveal width using Spectralis SD-OCT segmentation software in a young, healthy cohort. Thickness of all individual retinal layers were symmetrical at 2°, but not at 5° eccentricity away from the fovea.     

Interobserver Agreement in Detecting Spectral-Domain Optical Coherence Tomography Features of Diabetic Macular Edema

PurposeTo evaluate interobserver agreement for the detection of spectral-domain optical coherence tomography (SDOCT) features of diabetic macular edema (DME).MethodCross-sectional study in which 2 retinal specialists evaluated SDOCT scans from eyes receiving treatment for DME. Scans from 50 eyes with DME of 39 patients were graded for features of DME including intra-retinal fluid (IRF), diffuse retinal oedema (DRE), hyper-reflective foci (HRF), subretinal fluid (SRF), macular fluid and vitreomacular traction (VMT). Features were graded as present or absent at zones involving the fovea, 1mm from the fovea and the whole scan of 49 line scans. Analysis was performed using cross-tabulations for percentage concordance and kappa values (κ).ResultsIn the 2950 line scans analysed, there was an increase in percentage concordance for DRE and HRF when moving from a foveal line scan, 1mm zone and then to a whole scan analysis (88% vs 94% vs 96%) and (88% vs 94% vs 94%) respectively with κ ranging from substantial to almost perfect. Percentage concordance for SRF was 96% at all 3 regions analysed, whilst IRF was 96% at fovea and 98% at higher number of line-scans analysed. Concordance for MF was 100% at fovea and 98% at 1mm zone and whole scan with almost perfect and substantial κ respectively. κ agreement was substantial for VMT at all regions analysed.ConclusionWe report a high level of interobserver agreement in the detection of SDOCT features of DME. This finding is important as detection of macular fluid is used to guide retreatment with anti-angiogenic agents.     

Intravascular Optical Coherence Tomography Image Segmentation Based on Support Vector Machine Algorithm

Intravascular optical coherence tomography (IVOCT) is becoming more and more popular in clinical diagnosis of coronary atherosclerotic. However, reading IVOCT images is of large amount of work. This article describes a method based on image feature extraction and support vector machine (SVM) to achieve semi-automatic segmentation of IVOCT images. The image features utilized in this work including light attenuation coefficients and image textures based on gray level co-occurrence matrix. Different sets of hyper-parameters and image features were tested. This method achieved an accuracy of 83% on the test images. Single class accuracy of 89% for fibrous, 79.3% for calcification and 86.5% lipid tissue. The results show that this method can be a considerable way for semi-automatic segmentation of atherosclerotic plaque components in clinical IVOCT images.     

Vitreous Hyper-Reflective Dots in Optical Coherence Tomography and Cystoid Macular Edema after Uneventful Phacoemulsification Surgery

Purpose

To report the observation of hyper-reflective dots in the vitreous cavity using spectral domain optical coherence tomography (SD-OCT) after uneventful phacoemulsification cataract surgery and to investigate their association with cystoid macular edema (CME).

Materials and Methods

Medical records of consecutive Asian patients who had no preoperative retinopathy and underwent uneventful phacoemulsification cataract surgery from March 2012 through February 2013 were reviewed. SD-OCTs were performed before, 1 week, and 1 month after surgery. The number of vitreous hyper-reflective dots (VHDs) was counted in 5 OCT images of high-definition 5-line raster scans. The development of CME was assessed using postoperative 1-month OCT.

Results

In 74 eyes of 74 patients, all of three SD-OCTs with a signal to noise ratio of 0.6 or more were available and were analyzed in this study. In preoperative OCT, the VHD was observed in 2 (2.7%) of 74 eyes; one eye had 1 VHD and the other eye had 2 VHDs. In 72 eyes with no preoperative VHD, VHDs were observed in 40 (55.6%) eyes at 1 week after the surgery. In the multivariate analysis, the number of VHDs measured at 1 week after the surgery was significantly associated with CME development at 1 month after the surgery (odds ratio = 1.93, 95% confidence interval = 1.15 to 3.24, P = 0.012).

Conclusions

VHDs were frequently observed in OCT after uneventful phacoemulsification cataract surgery. VHDs observed at 1 week after the surgery may be a risk factor for the development of pseudophakic CME. Further studies are needed to identify the source of the VHDs.     

Macular Ganglion Cell Analysis Determined by Cirrus HD Optical Coherence Tomography for Early Detecting Chiasmal Compression

PurposeTo evaluate the performance of macular ganglion cell-inner plexiform layer (mGCIPL) measurement with Cirrus high-definition (HD) optical coherence tomography (OCT) for early detection of optic chiasmal compression.MethodsForty-six eyes of 46 patients with optic chiasmal compression caused by a pituitary adenoma (PA group), 31 eyes of 31 patients with normal tension glaucoma (NTG group), and 32 eyes of 32 normal participants (control group) were enrolled. The PA group was subdivided into two subgroups, which comprised patients with temporal visual field (VF) defects (perimetric PA group, 34 eyes) and without VF defect (preperimetric PA group, 12 eyes). The mGCIPL thickness and circumpapillary retinal nerve fiber layer (cpRNFL) thickness were measured using Cirrus HD-OCT. We calculated the number of patients who had an abnormal GCA sector map, defined as at least one yellow or red sector.ResultsEyes in the perimetric PA group had significantly decreased mGCIPL thickness in all sectors. Eyes in the preperimetric PA group had significantly thinner mGCIPL in the superior, superonasal, inferonasal, and inferior sectors than eyes in control group, but no changes in cpRNFL parameters were observed. The mGCIPL thickness in inferonasal area showed the greatest AUC value (0.965), followed by the superonasal area (0.958) for discriminating preperimetric PA group from the control group. A higher reduction rate of mGCIPL thickness was noted in the nasal sector compared to other sectors, which was irrespective of temporal visual field defects. The mGCIPL thickness maps showed superonasal (P = 0.003) and inferonasal thinning in the PA group (P = 0.003), while inferotemporal thinning was revealed in the NTG group (P = 0.001).ConclusionsMacular GCIPL thickness parameters obtained with the Cirrus HD-OCT were useful in early detection of chiasmal compression and differentiating from NTG by characteristic nasal mGCIPL thinning.     

Doppler Optical Coherence Tomography of Retinal Circulation

Noncontact retinal blood flow measurements are performed with a Fourier domain optical coherence tomography (OCT) system using a circumpapillary double circular scan (CDCS) that scans around the optic nerve head at 3.40 mm and 3.75 mm diameters. The double concentric circles are performed 6 times consecutively over 2 sec. The CDCS scan is saved with Doppler shift information from which flow can be calculated. The standard clinical protocol calls for 3 CDCS scans made with the OCT beam passing through the superonasal edge of the pupil and 3 CDCS scan through the inferonal pupil. This double-angle protocol ensures that acceptable Doppler angle is obtained on each retinal branch vessel in at least 1 scan. The CDCS scan data, a 3-dimensional volumetric OCT scan of the optic disc scan, and a color photograph of the optic disc are used together to obtain retinal blood flow measurement on an eye. We have developed a blood flow measurement software called "Doppler optical coherence tomography of retinal circulation" (DOCTORC). This semi-automated software is used to measure total retinal blood flow, vessel cross section area, and average blood velocity. The flow of each vessel is calculated from the Doppler shift in the vessel cross-sectional area and the Doppler angle between the vessel and the OCT beam. Total retinal blood flow measurement is summed from the veins around the optic disc. The results obtained at our Doppler OCT reading center showed good reproducibility between graders and methods (<10%). Total retinal blood flow could be useful in the management of glaucoma, other retinal diseases, and retinal diseases. In glaucoma patients, OCT retinal blood flow measurement was highly correlated with visual field loss (R²>0.57 with visual field pattern deviation). Doppler OCT is a new method to perform rapid, noncontact, and repeatable measurement of total retinal blood flow using widely available Fourier-domain OCT instrumentation. This new technology may improve the practicality of making these measurements in clinical studies and routine clinical practice.     

光学相干层析成像技术在内窥镜中的应用

光学相干层析成像(optical coherence tomography,OCT)技术是继X射线成像、核磁共振成像、超声成像等之后的一种新型的成像技术,其可光纤化的特点使得它易于医用电子内窥镜相结合。OCT内窥镜技术可实现对人体内部器官的高速率、高分辨率、无损伤、实时成像。主要介绍了OCT技术的种类、基本原理以及包括探测深度和纵向分辨率等的参数;简述了OCT内窥镜的发展历史以及最新成果,重点分析了光源对OCT内窥镜的影响。总结了OCT内窥镜的主要应用。     

光学相干层析用于牙齿病变的检测

阐述了适用于牙齿结构成像的光学相干层析成像(OCT)系统。系统光源中心波长为1 310 nm,成像分辨率10μm,在牙内成像深度2 mm,成像速度1幅/秒,系统信噪比100 dB。利用此OCT仪清晰检测到牙齿样品的牙釉质和牙本质,观察到牙釉质与牙本质的分界面以及正常牙齿牙釉质与龋齿牙齿牙釉质OCT图像的区别。进一步设计研制了适用于口腔内探测的小型OCT探头。     

Visualization of Plant Tissues by Optical Coherence Tomography

The internal structure of plant tissues was visualized with optical coherence tomography (OCT). This noninvasive method is suitable for examining intact plants; it produces two-dimensional images of plant tissues at a penetration depth of 1–2 mm from the surface. The potential use of OCT was assessed on Tradescantia blossfeldiana Mild. Plant tissue images measuring 1.5 × 2 mm were obtained in vivo with a spatial resolution of 15 m. The radiation power incident on a sample was 0.5 mW. The acquisition of a two-dimensional image consisting of 200 × 200 pixels required 1–3 s. The OCT method can be used to visualize not only plant tissues and tissue boundaries but also the structure of individual cells.     

光学相干层析分子成像方法初探

介绍了分子对比剂在光学相干层析成像(optical coherence tom ography,OCT)技术中的研究现状,概述了迄今出现的几种不同的光学相干层析分子成像方法(m olecu lar contrast OCT,简称为MCOCT),讨论了MCOCT的几个重要的实际问题:对比剂的选择范围、激发光强的限制、各种方法灵敏度比较以及MCOCT应用于临床与生物学领域需要考虑的因素。     

Keratometric Index Obtained by Fourier-Domain Optical Coherence Tomography

Purpose

To determine the keratometric indices calculated based on parameters obtained by Fourier-domain optical coherence tomography (FD-OCT).

Methods

The ratio of anterior corneal curvature to posterior corneal curvature (Ratio) and keratometric index (N) were calculated within central 3 mm zone with the RTVue FD-OCT (RTVue, Optovue, Inc.) in 186 untreated eyes, 60 post-LASIK/PRK eyes, and 39 keratoconus eyes. The total corneal powers were calculated using different keratometric indices: K_cal based on the mean calculated keratometric index, K_1.3315 calculated by the keratometric index of 1.3315, and K_1.3375 calculated by the keratometric index of 1.3375. In addition, the total corneal powers based on Gaussian optics formula (K_actual) were calculated.

Results

The means for Ratio in untreated controls, post-LASIK/PRK group and keratoconus group were 1.176 ± 0.022 (95% confidence interval (CI), 1.172–1.179), 1.314 ± 0.042 (95%CI, 1.303–1.325) and 1.229 ± 0.118 (95%CI, 1.191–1.267), respectively. And the mean calculated keratometric index in untreated controls, post-LASIK/PRK group and keratoconus group were 1.3299 ± 0.00085 (95%CI, 1.3272–1.3308), 1.3242 ± 0.00171 (95%CI, 1.3238–1.3246) and 1.3277 ± 0.0046 (95%CI, 1.3263–1.3292), respectively. All the parameters were normally distributed. The differences between K_cal and K_actual, K_1.3315 and K_actual, and K_1.3375 and K_actual were 0.00 ± 0.11 D, 0.21 ± 0.11 D and 0.99 ± 0.12 D, respectively, in untreated controls; -0.01 ± 0.20 D, 0.85 ± 0.18 D and 1.56 ± 0.16 D, respectively, in post-LASIK/PRK group; and 0.03 ± 0.67 D, 0.56 ± 0.70 D and 1.40 ± 0.76 D, respectively, in keratoconus group.

Conclusion

The calculated keratometric index is negatively related to the ratio of anterior corneal curvature to posterior corneal curvature in untreated, post-LASIK/PRK, and keratoconus eyes, respectively. Using the calculated keratometric index may improve the prediction accuracies of total corneal powers in untreated controls, but not in post-LASIK/PRK and keratoconus eyes.     

Integrated Photoacoustic Ophthalmoscopy and Spectral-domain Optical Coherence Tomography

Both the clinical diagnosis and fundamental investigation of major ocular diseases greatly benefit from various non-invasive ophthalmic imaging technologies. Existing retinal imaging modalities, such as fundus photography¹, confocal scanning laser ophthalmoscopy (cSLO)², and optical coherence tomography (OCT)³, have significant contributions in monitoring disease onsets and progressions, and developing new therapeutic strategies. However, they predominantly rely on the back-reflected photons from the retina. As a consequence, the optical absorption properties of the retina, which are usually strongly associated with retinal pathophysiology status, are inaccessible by the traditional imaging technologies.Photoacoustic ophthalmoscopy (PAOM) is an emerging retinal imaging modality that permits the detection of the optical absorption contrasts in the eye with a high sensitivity^4-7 . In PAOM nanosecond laser pulses are delivered through the pupil and scanned across the posterior eye to induce photoacoustic (PA) signals, which are detected by an unfocused ultrasonic transducer attached to the eyelid. Because of the strong optical absorption of hemoglobin and melanin, PAOM is capable of non-invasively imaging the retinal and choroidal vasculatures, and the retinal pigment epithelium (RPE) melanin at high contrasts ^6,7. More importantly, based on the well-developed spectroscopic photoacoustic imaging^5,8 , PAOM has the potential to map the hemoglobin oxygen saturation in retinal vessels, which can be critical in studying the physiology and pathology of several blinding diseases ⁹ such as diabetic retinopathy and neovascular age-related macular degeneration.Moreover, being the only existing optical-absorption-based ophthalmic imaging modality, PAOM can be integrated with well-established clinical ophthalmic imaging techniques to achieve more comprehensive anatomic and functional evaluations of the eye based on multiple optical contrasts^6,10 . In this work, we integrate PAOM and spectral-domain OCT (SD-OCT) for simultaneously in vivo retinal imaging of rat, where both optical absorption and scattering properties of the retina are revealed. The system configuration, system alignment and imaging acquisition are presented.     

Optical Coherence Tomography for live imaging of mammalian development

Understanding the nature and mechanism of congenital defects of the different organ systems in humans has heavily relied on the analysis of the corresponding mutant phenotypes in rodent models. Optical Coherence Tomography (OCT) has recently emerged as a powerful tool to study early embryonic development. This non-invasive optical methodology does not require labeling and allows visualization of embryonic tissues with single cell resolution. Here, we will discuss how OCT can be applied for structural imaging of early mouse and rat embryos in static culture, cardiodynamic and blood flow analysis, and in utero embryonic imaging at later stages of gestation, demonstrating how OCT can be used to assess structural and functional birth defects in mammalian models.     

Optical Coherence Tomography and Autofluorescence Imaging of Human Tonsil

For the first time, we present co-registered autofluorescence imaging and optical coherence tomography (AF/OCT) of excised human palatine tonsils to evaluate the capabilities of OCT to visualize tonsil tissue components. Despite limited penetration depth, OCT can provide detailed structural information about tonsil tissue with much higher resolution than that of computed tomography, magnetic resonance imaging, and Ultrasound. Different tonsil tissue components such as epithelium, dense connective tissue, lymphoid nodules, and crypts can be visualized by OCT. The co-registered AF imaging can provide matching biochemical information. AF/OCT scans may provide a non-invasive tool for detecting tonsillar cancers and for studying the natural history of their development.     

Study of the Morphological and Functional State of Higher Plant Tissues by Optical Coherence Microscopy and Optical Coherence Tomography

Comparative analysis of two optical methods—optical coherence tomography (OCT) and optical coherence microscopy (OCM)—was made for vital visualization of plant tissues in tomato (Lycopersicon esculentum Mill), spiderwort (Tradescantia pallida (Rose) D. Hunt), orach (Atriplex sp.), and leaves and seeds of medium starwort (Stellaria media L.). The obtained OCT- and OCM-images allowed the morphological and functional state of plant tissues to be assessed in vivo. A higher spatial resolution of the OCM method, as compared to OCT method, allowed plant morphological structures to be identified with greater confidence. The morphological and functional state of tissues can be monitored with a time resolution of 1–4 s in intact plants, without removing them from the habitat.__________Translated from Fiziologiya Rastenii, Vol. 52, No. 4, 2005, pp. 628–634.Original Russian Text Copyright © 2005 by Kutis, Sapozhnikova, Kuranov, Kamenskii.     

Tissue-Specific Optical Mapping Models of Swine Atria Informed by Optical Coherence Tomography

Computational models and experimental optical mapping of cardiac electrophysiology serve as powerful tools to investigate the underlying mechanisms of arrhythmias. Modeling can also aid the interpretation of optical mapping signals, which may have different characteristics with respect to the underlying electrophysiological signals they represent. However, despite the prevalence of atrial arrhythmias such as atrial fibrillation, models of optical electrical mapping incorporating realistic structure of the atria are lacking. Therefore, we developed image-based models of atrial tissue using structural information extracted from optical coherence tomography (OCT), which can provide volumetric tissue characteristics in high resolution. OCT volumetric data of four swine atrial tissue samples were used to develop models incorporating tissue geometry, tissue-specific myofiber orientation, and ablation lesion regions. We demonstrated the use of these models through electrophysiology and photon scattering simulations. Changes in transmural electrical conduction were observed with the inclusion of OCT-derived, depth-resolved fiber orientation. Additionally, the amplitude of optical mapping signals were not found to correspond with lesion transmurality because of lesion geometry and electrical propagation occurring beyond excitation light penetration. This work established a framework for the development of tissue-specific models of atrial tissue derived from OCT imaging data, which can be useful in future investigations of electrophysiology and optical mapping signals with respect to realistic atrial tissue structure.     

基于谱域相干的细胞动态检测技术

将谱域光学相干层析技术（Spectral-domain Optical Coherence Tomography,SD-OCT）应用于细胞膜厚的动态检测,引入了光谱干涉测量方法,通过对干涉光谱信号进行离散傅里叶变换得到表示物体深度位置信息的光程差信号。搭建了一套SD-OCT测量系统并设计了专门用于细胞检测的样品平台,检测精度达到μm量级,信噪比达70 db,是一种快速、实时、非接触的细胞分子膜厚动态检测技术。     

Age-Related Vitamin D Deficiency Is Associated with Reduced Macular Ganglion Cell Complex: A Cross-Sectional High-Definition Optical Coherence Tomography Study

Background

Vitamin D deficiency is associated with smaller volume of optic chiasm in older adults, indicating a possible loss of the visual axons and their cellular bodies. Our objective was to determine whether vitamin D deficiency in older adults is associated with reduced thickness of the ganglion cell complex(GCC) and of the retinal nerve fibre layer(RNFL), as measured with high-definition optical coherence tomography(HD-OCT).

Methods

Eighty-five French older community-dwellers without open-angle glaucoma and patent age-related macular degeneration(mean, 71.1±4.7years; 45.9%female) from the GAIT study were separated into 2 groups according to serum 25OHD level(i.e., deficient≤25nmol/L or sufficient>25nmol/L). Measurements of GCC and RNFL thickness were performed using HD-OCT. Age, gender, body mass index, number of comorbidities, dementia, functional autonomy, intracranial volume, visual acuity, serum calcium concentration and season of testing were considered as potential confounders.

Results

Mean serum 25OHD concentration was 58.4±26.8nmol/L. Mean logMAR visual acuity was 0.03±0.06. Mean visual field mean deviation was -1.25±2.29dB. Patients with vitamin D deficiency(n=11) had a reduced mean GCC thickness compared to those without vitamin D deficiency(72.1±7.4μm versus 77.5±7.5μm, P=0.028). There was no difference of the mean RNFL thickness in these two groups(P=0.133). After adjustment for potential confounders, vitamin D deficiency was associated with reduced GCC thickness(ß=-5.12, P=0.048) but not RNFL thickness(ß=-9.98, P=0.061). Specifically, vitamin D deficiency correlated with the superior medial GCC area(P=0.017) and superior temporal GCC area(P=0.010).

Conclusions

Vitamin D deficiency in older patients is associated with reduced mean GCC thickness, which can represent an early stage of optic nerve damage, prior to RNFL loss.