Sheng You

Plus disease diagnosis in retinopathy of prematurity: vascular tortuosity as a function of distance from optic disk.

Purpose: To examine vascular tortuosity as a function of distance from the optic disk in infants with retinopathy of prematurity. Methods: Thirty-four wide-angle retinal images from infants with retinopathy of prematurity were reviewed by 22 experts. A reference standard for each image was defined as the diagnosis (plus vs. not plus) given by the majority of experts. Tortuosity, defined as vessel length divided by straight line distance between vessel end points, was calculated as a function of distance from the disk margin for arteries and veins using computer-based methods developed by the authors. Results: Mean cumulative tortuosity increased with distance from the disk margin, both in 13 images with plus disease (P = 0.007 for arterial tortuosity [n = 62 arteries], P < 0.001 for venous tortuosity [n = 58 veins] based on slope of best fit line by regression), and in 21 images without plus disease (P < 0.001 for arterial tortuosity [n = 94 arteries], P <0 .001 for venous tortuosity [n = 85 veins]). Images with plus disease had significantly higher vascular tortuosity than images without plus disease (P < 0.05), up to 7.0 disk diameters from the optic disk margin. Conclusion: Vascular tortuosity was higher peripherally than centrally, both in images with and without plus disease, suggesting that peripheral retinal features may be relevant for retinopathy of prematurity diagnosis.

Analysis of Underlying Causes of Inter-expert Disagreement in Retinopathy of Prematurity Diagnosis: Application of Machine Learning Principles

OBJECTIVE Inter-expert variability in image-based clinical diagnosis has been demonstrated in many diseases including retinopathy of prematurity (ROP), which is a disease affecting low birth weight infants and is a major cause of childhood blindness. In order to better understand the underlying causes of variability among experts, we propose a method to quantify the variability of expert decisions and analyze the relationship between expert diagnoses and features computed from the images. Identification of these features is relevant for development of computer-based decision support systems and educational systems in ROP, and these methods may be applicable to other diseases where inter-expert variability is observed. METHODS The experiments were carried out on a dataset of 34 retinal images, each with diagnoses provided independently by 22 experts. Analysis was performed using concepts of Mutual Information (MI) and Kernel Density Estimation. A large set of structural features (a total of 66) were extracted from retinal images. Feature selection was utilized to identify the most important features that correlated to actual clinical decisions by the 22 study experts. The best three features for each observer were selected by an exhaustive search on all possible feature subsets and considering joint MI as a relevance criterion. We also compared our results with the results of Cohens Kappa [36] as an inter-rater reliability measure. RESULTS The results demonstrate that a group of observers (17 among 22) decide consistently with each other. Mean and second central moment of arteriolar tortuosity is among the reasons of disagreement between this group and the rest of the observers, meaning that the group of experts consider amount of tortuosity as well as the variation of tortuosity in the image. CONCLUSION Given a set of image-based features, the proposed analysis method can identify critical image-based features that lead to expert agreement and disagreement in diagnosis of ROP. Although tree-based features and various statistics such as central moment are not popular in the literature, our results suggest that they are important for diagnosis.

Principal Curved Based Retinal Vessel Segmentation towards Diagnosis of Retinal Diseases

The extraction of retinal vessels plays an important role in the diagnosis and study of retinal diseases, such as Age-related Macular Degeneration (AMD), Diabetic Retinopathy, Retinopathy of Prematurity (ROP). Vessel diameters, tortuosity, branch lengths, angles, and bifurcations are essential to diagnosing these diseases. However, this is a challenging task due to high noise levels, the low contrast of thin vessels to the background, non-uniform illumination, and the central light reflex. Our goal here is to develop a framework to accurately segment the retinal vessels as a preprocessing step for the feature extraction of the vessels towards the future disease diagnosis. In this paper, we present a principal curve based retinal vessel segmentation approach to achieve this goal. We first use the isotropic Gaussian kernel Frangi filter to enhance the retinal vessels and measure the diameters of them. A multiscale principal curve projection and tracing algorithm is then proposed to identify the centerlines of the vessels in the output image of the Franfi filter using the underlying kernel smoothing interpolation of the intensities. The estimated vessel radius from the Frangi filter are used as the bandwidth of the kernel interpolation in the principal curve projection and tracing step. The vessel features toward diagnosing and analyzing the diseases can be extracted from our segmentation results. The presented approach is implemented on a publicly available DRIVE database [16].

Observer and feature analysis on diagnosis of retinopathy of prematurity

Retinopathy of prematurity (ROP) is a disease affecting low-birth weight infants and is a major cause of childhood blindness. However, human diagnoses is often subjective and qualitative. We propose a method to analyze the variability of expert decisions and the relationship between the expert diagnoses and features. The analysis is based on Mutual Information and Kernel Density Estimation on features. The experiments are carried out on a dataset of 34 retinal images diagnosed by 22 experts. The results show that a group of observers decide consistently with each other and there are popular features that have a high correlation with labels.

A novel line detection method in space-time images for microvascular blood flow analysis in sublingual microcirculatory videos

Recent evidence suggests that quantitative assessment of microcirculatory dysfunction may indicate certain disease states [1, 2, 3]. Relevant microcirculatory hemodynamic parameters include total vessel density, density of perfused vessels, proportion of perfused vessels, and perfusion heterogeneity index. In one non-invasive, clinical approach, a handheld video microscope placed under the tongue records images of blood flow in small (<; 20μm) and medium (approximately 20-100μm) diameter vessels. Hemodynamic parameters are computed from measurements of vessel geometry and blood flow rates. Current technology is limited by poor dynamic range, low resolution, poor image stability, and pressure artifacts. Video images are analyzed quantitatively and semi-quantitatively by trained image analysts using a time-consuming, semi-automated techniques for vessel segmentation, and blood flow measurements. Space-time images are generated for quantitative velocity estimation. We propose a novel line detection method to automatically estimate the orientation of red blood cell (RBC) or plasma gap traces in space-time images. Velocities of RBCs can then be calculated based on the estimated orientation. The proposed automated method for velocity estimation was implemented for 80 vessels and compared with visual estimation of reference slope in space-time diagrams by a trained image analyst. Finally, the proposed method is compared with a Hough transform based velocity estimation method.

Towards respiration management in radiation treatment of lung tumors: Transferring regions of interest from planning CT to kilovoltage X-ray images

Tracking of lung tumors is imperative for improved radiotherapy treatment. However, the motion of the thoracic organs makes it a complicated task. 4D CT images acquired prior to treatment provide valuable information regarding the motion of organs and tumor, since it is manually annotated. In order to track tumors using treatment-day X-ray images (kV images), we need to find the correspondence with CT images so that projection of tumor region of interest will provide a good estimate about the position of the tumor on the X-ray image. In this study, we propose a method to estimate the alignment and respiration phase corresponding to X-ray images using 4D CT data. Our approach generates Digitally Reconstructed Radiographs (DRRs) using bilateral filter smoothing and computes rigid registration with kV images since the position and orientation of patient might differ between CT and treatment-day image acquisition processes. Instead of using landmark points, our registration method makes use of Kernel Density Estimation over the edges that are not affected much by respiration. To estimate the phase of X-ray, we apply template matching techniques between the lung regions of X-ray and registered DRRs. Our approach gives accurate results for rigid registration and provides a starting point to track tumors using the X-ray images during the treatment.

Extraction of samples from airway and vessel trees in 3D lung CT based on a multi-scale principal curve tracing algorithm

The extraction of airway and vessel trees plays an important role in the diagnosis and treatment planning of lung diseases. However, this is a challenging task due to the small size of the anatomical structures, noise, or artifacts in the image. The similar intensity values between the lung parenchyma and airway lumen, the airway wall and the blood vessels make extraction particularly difficult. Our method detailed herein presents an automatic extraction of samples of both the airways and vessels from the three-dimensional computed tomography (3D-CT) based on the multi-scale principal curve algorithm. The image is first thresholded to find airway or vessel candidates according to their corresponding Hounsfield units (HU). The Frangi filter is then used to extract the tubular structures and remove background noise. Finally, a multi-scale principal curve projection and tracing algorithm is applied on the filtered image to identify the centerlines of the airway and vessel trees.

Microvascular blood flow estimation in sublingual microcirculation videos based on a principal curve tracing algorithm

Microcirculatory perfusion is an important metric for diagnosing pathological conditions in patients. Capillary density and red blood cell (RBC) velocity provide a measure of tissue perfusion. Estimating RBC velocity is a challenging problem due to noisy video sequences, low contrast between the vessels and the background, and thousands of RBCs moving rapidly through video sequences. Typically, physicians manually trace small blood vessels and visually estimate RBC velocities. The task is labor intensive, tedious, and time-consuming. In this paper, we present a novel application of a principal curve tracing algorithm to automatically track RBCs across video frames and estimate their velocity based on the displacements of RBCs between two consecutive frames. The proposed method is implemented in one sublingual microcirculation video of a healthy subject.

A Novel Application of Principal Surfaces to Segmentation in 4D-CT for Radiation Treatment Planning

Radiation therapy is one of the most effective options used in the treatment of about half of all people with cancer. A critical goal in radiation therapy is to deliver optimal radiation doses to the observed tumor while sparing the surrounding healthy tissues. Radiation oncologists typically manually delineate normal and diseased structures on three-dimensional computed tomography~(3D-CT) scans. Manual delineation is a labor intensive, tedious and time-consuming task. In recent years, concerns about respiration induced motion have led to the popularity of four-dimensional computed tomography~(4D-CT) for the tracking of tumors and deformation of organs. However, as manually contouring in all phases would be prohibitively expensive, the development of fast, robust, and automatic segmentation tools has been an active area of research in 4D radiotherapy. In this paper, we describe a novel application of principal surfaces for the propagation of contours in 4D-CT studies. Regions of interest~(ROIs) are manually delineated slice-by-slice in the reference 3D-CT scans. Edges are detected on all of the slices of the target 3D-CT phase. A kernel density estimation~(KDE) based on the detected edges is then calculated. The principal surface algorithm is applied to find the ridges of the edge KDE to provide the object contours. Manually drawn contours from the reference phase are used as an initialization. Contours of ROIs are propagated recursively in all consecutive phases to complete a respiration cycle. Results are provided for a phantom data set of simulated tumor motion as well as on a de-identified data set of the lung of a patient. Evaluation of the efficacy of automatic segmentation in organs and tumors are based on the comparison between manually drawn contours and automatically delineated contours. The Dice coefficients are approximately 0.97 for the lung tumor on the phantom data sets and 0.95 for the patient data sets. The centroid distances between manually delineated lung volume and automatically segmented lung volume in each CT direction are

Principal curve based semi-automatic segmentation of organs in 3D-CT

Radiation therapy plays an important and effective role in the treatment of cancer. A main goal in radiation therapy is to deliver high radiation doses to the perceived tumors while minimizing radiation to surrounding normal tissues. Manual delineation of tumors and organs-at-risk(OARs) on three-dimensional computed tomography (3D-CT) is both a time-consuming and labor intensive task, and there maybe variability between manual delineations by different radiation oncologists. In this paper, we present a semi-supervised method to segment the contours of organs represented by piecewise linear segments connected with a small number of points given the users input in one or more slices as an approximate initialization. This method detects ridge samples from the kernel interpolation of the edge map and approximates the shape of organs using piecewise linear segments among those sample points based on the principal curve score. Results are provided in two 3D-CT scans. Evaluation of the efficacy of our semiautomatic segmentation method is based on the overlapping ratio between the manually delineated contours and the semiautomatic segmented contours represented by a small number of points. The preserved points can be as low as 10 percent of the initial manual points, and the Dice Coefficients are approximately 0.93 for lung segmentation.