Risk prediction of chronic diseases with a two-stage semi-supervised clustering method.

Early detection of chronic diseases such as cardiovascular disease (CVD) and diabetes can make the difference between life and death. Previous studies have demonstrated the feasibility of disease diagnosis and prediction using machine learning and disease-indicating biomarkers. The aim of this study is to develop a method to detect the risk of future disease even when disease-indicating biomarker readings are in the normal range. Data from the US Centers for Disease Control and Prevention (CDC) National Health and Nutrition Examination Surveys (NHANES) are used for this study. A two-stage semi-supervised K-Means (SSK-Means) clustering approach was developed to identify the underlying risk of each individual and categorize them into high or low-risk groups for CVD and diabetes. Our developed method of classification can identify groups as high risk or low risk, even if they would have been considered normal using traditional biomarker threshold criteria. For CVD, the SSK-Means clustering results showed that individuals over 30 years of age in the high-risk group were almost twice as likely to develop CVD as individuals in the low-risk group. For diabetes, the SSK-Means clustering results showed that individuals over 50 years in the high-risk group have at least two times the risk of developing diabetes compared with individuals in the low-risk group.

Retinal nerve fiber layer retardation measurements using a polarization-sensitive fundus camera.

To measure the retardation distribution of the optic retinal nerve fiber layer (RNFL) from a single image, we have developed a new polarization analysis system that is able to detect the Stokes vector using a fundus camera. The polarization analysis system is constructed with a CCD area image sensor, a linear polarizing plate, a microphase plate array, and a circularly polarized light illumination unit. In this system, the Stokes vector expressing the whole state of polarization is detected, and the influence of the background scattering in the retina and of the retardation caused by the cornea are numerically eliminated. The measurement method is based on the hypothesis that the retardation process of the eye optics can be quantified by a numerical equation that consists of a retardation matrix of all the polarization components. We show the method and the measurement results for normal eyes. Our results indicate that the present method may provide a useful means for the evaluation of retardation distribution of the RNFL.

Automated layer segmentation of macular OCT images using dual-scale gradient information.

A novel automated boundary segmentation algorithm is proposed for fast and reliable quantification of nine intra-retinal boundaries in optical coherence tomography (OCT) images. The algorithm employs a two-step segmentation schema based on gradient information in dual scales, utilizing local and complementary global gradient information simultaneously. A shortest path search is applied to optimize the edge selection. The segmentation algorithm was validated with independent manual segmentation and a reproducibility study. It demonstrates high accuracy and reproducibility in segmenting normal 3D OCT volumes. The execution time is about 16 seconds per volume (480x512x128 voxels). The algorithm shows potential for quantifying images from diseased retinas as well.

Clinical evaluation of the Topcon BV-1000 automated subjective refraction system.

BACKGROUND: The BV-1000 is a new instrument that performs binocular autorefraction and subjective refraction. The aim of the study was to evaluate the accuracy of the BV-1000 compared with subjective refraction without cycloplegia and the repeatability of the BV-1000 in subjects without cycloplegia. METHODS: BV-1000 binocular autorefraction (mean of three measures) and automated subjective refraction was performed on 100 eyes and compared with practitioner subjective refraction. After BV-1000 assessment, the result was modified by the investigator in the form of a binocular addition; the practitioner was masked as to the result of the BV-1000. Repeatability was evaluated by comparing the test-retest data on 40 eyes. Refraction data were converted into Fourier component terms: mean spherical equivalent (MSE), J0, J45, and vector dioptric difference (VDD). RESULTS: Mean difference of MSE (BV-1000 subjective vs. practitioner) was 0.05 +/- 0.35 D (p = 0.13), whereas after binocular addition, the mean difference in MSE was -0.01 +/- 0.21 D (p = 0.67). The mean difference of MSE (BV-1000 objective vs. practitioner) was +0.10 +/- 0.37 D (p < 0.05). Mean difference in J0 and J45 was -0.03 +/- 0.12 D (p = 0.09) and -0.02 +/- 0.10 D (p = 0.06), respectively. More than 95% of cylinder powers were within +/- 0.50 D. The median VDD was 0.35 D (range, 1.38 D; interquartile range, 0.31) for comparisons of BV-1000 subjective refraction and the practitioner. A median VDD of 0.25 D (range, 1.06 D; interquartile range, 0.16) was found after binocular addition correction. There was no statistically significant difference (p = 0.73) in spherical intereye difference between BV-1000 subjective and subjective refraction. Mean spherical intereye difference (spherical balancing error) between the BV-1000 and practitioner subjective refraction was -0.02 +/- 0.24 D. Repeatability as described by the mean difference between occasions for MSE was found to be -0.06 +/- 0.19 D, -0.02 +/- 0.08 D for J0, and -0.03 +/- 0.07 D for J45. CONCLUSIONS: The subjective refraction routine within the BV-1000 improves the results of simultaneous binocular autorefraction data. The BV-1000 provides an accurate and repeatable measure of subjective refraction after binocular addition.