Hyunjoong Kim

Classification Trees With Unbiased Multiway Splits

Two univariate split methods and one linear combination split method are proposed for the construction of classification trees with multiway splits. Examples are given where the trees are more compact and hence easier to interpret than binary trees. A major strength of the univariate split methods is that they have negligible bias in variable selection, both when the variables differ in the number of splits they offer and when they differ in the number of missing values. This is an advantage because inferences from the tree structures can be adversely affected by selection bias. The new methods are shown to be highly competitive in terms of computational speed and classification accuracy of future observations.

Reversal of Lamina Cribrosa Displacement after Intraocular Pressure Reduction in Open-Angle Glaucoma

PURPOSE To compare the change in lamina cribrosa (LC) displacement in response to intraocular pressure (IOP) lowering in patients with open-angle glaucoma (OAG) using spectral-domain optical coherence tomography (SD-OCT). DESIGN Observational case series. PARTICIPANTS One hundred eyes of 100 patients with OAG in whom IOP at the follow-up examination had decreased by at least 20% compared with the baseline IOP. METHODS Serial horizontal B-scan images of the optic nerve head (ONH) were obtained from each eye using enhanced depth imaging SD-OCT. Approximately 65 B-scans covering the optic discs were obtained before and 3 to 6 months after lowering IOP. The baseline and follow-up LC depths (the distance from the Bruchs membrane opening plane to the level of the anterior LC surface) were measured in B-scan images from each eye. MAIN OUTCOME MEASURES The mean and maximum amount of reductions in LC depth measured in the 7 selected B-scan images. RESULTS Intraocular pressure decreased from 21.2±9.1 to 10.5±2.6 mmHg. The percent of IOP reduction was significantly related to the untreated IOP (P < 0.001). There was a significant decrease in the LC depth at the follow-up examination compared with the baseline value (P < 0.001). The magnitude of LC depth reduction was significantly associated with younger age, higher untreated IOP, higher baseline IOP, and greater percent of IOP reduction (all P < 0.02). CONCLUSIONS Reversal of the LC displacement was observed after IOP-lowering treatment in OAG. The degree of LC displacement reversal was related to the amount of IOP lowering.

Ability of Stratus OCT to detect progressive retinal nerve fiber layer atrophy in glaucoma.

PURPOSE To evaluate the ability of Stratus optical coherence tomography (OCT version 4.0; Carl Zeiss Meditec, Inc., Dublin, CA) to detect progressive glaucomatous retinal nerve fiber layer (RNFL) atrophy observed by red-free RNFL photography. METHODS Intersession test-retest variability of each clock hour, quadrant, and average RNFL thickness was determined in 53 control subjects. The sensitivity and specificity of OCT for identification of progressive RNFL atrophy were tested on subjects in whom this condition was clearly observed in red-free RNFL photographs (n=27) and in another control group (n=62), according to criteria derived from test-retest variability. RESULTS The sensitivity of Stratus OCT RNFL measurement ranged from 14.8% (for average RNFL thickness) to 85.2% (for clock hour thickness) when tested at the 95% confidence level. The specificity of Stratus OCT RNFL measurement was approximately 95% for average RNFL thickness, but decreased considerably with clock hour (59.7%) and quadrant thickness (77.4%). This is presumably because multiple testing was used for multiple clock hours and quadrants. When calculated based on two consecutive follow-up examinations, the specificity for the clock hour measurements increased to 86.6% and that for quadrant thickness increased to 92.5%. The OCT-measured RNFL thickness change showed excellent topographic agreement with the progressive RNFL atrophy observed using RNFL photography. CONCLUSIONS Within the limits of retest variability, Stratus OCT detects progressive RNFL atrophy with high sensitivity and moderate specificity in cases showing localized progressive loss of retinal nerve fibers in red-free photographs. The specificity can be improved by use of multiple measurements. Stratus OCT is a potentially useful technique for detection of glaucoma progression.

Classification Trees With Bivariate Linear Discriminant Node Models

This article introduces a classification tree algorithm that can simultaneously reduce tree size, improve class prediction, and enhance data visualization. We accomplish this by fitting a bivariate linear discriminant model to the data in each node. Standard algorithms can produce fairly large tree structures because they employ a very simple node model, wherein the entire partition associated with a node is assigned to one class. We reduce the size of our trees by letting the discriminant models share part of the data complexity. Being themselves classifiers, the discriminant models can also help to improve prediction accuracy. Finally, because the discriminant models use only two predictor variables at a time, their effects are easily visualized by means of two-dimensional plots. Our algorithm does not simply fit discriminant models to the terminal nodes of a pruned tree, as this does not reduce the size of the tree. Instead, discriminant modeling is carried out in all phases of tree growth and the misclassification costs of the node models are explicitly used to prune the tree. Our algorithm is also distinct from the “linear combination split” algorithms that partition the data space with arbitrarily oriented hyperplanes. We use axis-orthogonal splits to preserve the interpretability of the tree structures. An extensive empirical study with real datasets shows that, in general, our algorithm has better prediction power than many other tree or nontree algorithms.

Microstructure of β-Zone Parapapillary Atrophy and Rate of Retinal Nerve Fiber Layer Thinning in Primary Open-Angle Glaucoma

OBJECTIVE To investigate whether the rate of retinal nerve fiber layer (RNFL) thinning differs according to the microstructure of β-zone parapapillary atrophy (PPA) as evaluated by spectral-domain (SD) optical coherence tomography (OCT) in primary open-angle glaucoma (POAG). DESIGN Prospective, observational, comparative study. PARTICIPANTS A total of 195 eyes with POAG that had been evaluated by serial SD-OCT RNFL thickness measurements for ≥ 2 years. METHODS On the basis of the extent of Bruchs membrane (BM) within the β-zone PPA (area without retinal pigment epithelium [RPE]), as shown in the infrared fundus images, PPA was divided into PPA+BM (PPA with intact BM) and PPA-BM (PPA devoid of BM). Eyes were categorized into group A (having PPA+BM only, n=64), group B (having both PPA+BM and PPA-BM, n=58), group C (having PPA-BM only, n=32), and group D (without β-zone PPA, n=41). The rate of progressive OCT RNFL thinning was determined by linear regression and compared between groups. Factors influencing the rate of RNFL thinning were evaluated, including age, sex, follow-up duration, history of filtering surgery, baseline RNFL thickness, baseline intraocular pressure (IOP), mean IOP and IOP fluctuation during follow-up, PPA types, baseline PPA width, PPA width increase, axial length (AXL), central corneal thickness, and visual field mean deviation (MD). MAIN OUTCOME MEASURES Rate of thinning of OCT RNFL thicknesses over time. RESULTS Patients in groups B and C were significantly younger and more myopic, and had a greater AXL, than those in groups A and D (all P<0.001). The rate of global RNFL thinning was significantly faster in group A (-1.66 ± 2.94 μm/year) than in the other groups (group B, -0.87 ± 1.28 μm/year; group C, 0.20 ± 1.86 μm/year; group D, -0.28 ± 1.74 μm/year; P = 0.001). Multivariate regression showed a significant association of shorter follow-up period (P = 0.016), greater baseline global RNFL thickness (P = 0.035), type of β-zone PPA (group A, P = 0.023), and greater baseline PPA+BM width (P = 0.034) with a faster rate of RNFL thinning. CONCLUSIONS The rate of RNFL thinning differed according to the microstructure of β-zone PPA. It was faster for eyes with β-zone PPA with intact BM than for eyes without β-zone PPA or with β-zone PPA devoid of BM.

Energy-Saving Techniques for Reducing CO 2 Emissions in Elementary Schools

Global warming has grown into a significantly intense global issue, prompting all nations to exert various efforts to reduce their greenhouse gas emission, according to the United Nations Framework Convention on Climate Change (UNFCCC). The South Korean government is likewise working on reducing its greenhouse gas emission and, as part of its effort, it is currently promoting the Green-School Project. This project aims to create energy-saving schools by renovating deteriorated school buildings. In this study, CO2 emission reduction scenarios are sought by applying energy-saving techniques (ESTs) to educational facilities, and an optimal scenario is created by performing life cycle cost (LCC) analysis for each scenario. To this end, the amounts of energy consumption and CO2 emission reduction in two schools were evaluated using DesignBuilder simulation, based on 15 scenarios with combinations of four ESTs. LCC analysis was also performed. In terms of energy efficiency, CO2 emission reduction, and LCC analy...

Correlation between topographic profiles of localized retinal nerve fiber layer defects as determined by optical coherence tomography and red-free fundus photography.

PurposeTo evaluate the topographic relationship of localized retinal nerve fiber layer (RNFL) defects as determined by optical coherence tomography (OCT) and by red-free fundus photography. MethodsSixty-five eyes of 65 patients with localized, wedge-shaped RNFL defects identified in red-free fundus photographs and with matching visual field defects were scanned with the Stratus OCT. On the resulting RNFL thickness report printout, segments of the line graph outside of the 95 and 99% normal limits were defined as “5% OCT defect” and “1% OCT defect,” respectively. Correlations between the angular location and angular width of RNFL defects as determined by OCT (OCT defects) and by red-free fundus photography (red-free defects) were evaluated. ResultsAngular locations and widths of both 5 and 1% OCT defects were significantly correlated with corresponding red-free defects (all P values<0.001). ConclusionsThere is a strong topographic correlation between characteristics of localized RNFL defects as determined by OCT and red-free photography.

A decision support model for improving a multi-family housing complex based on CO2 emission from electricity consumption

The number of deteriorated multi-family housing complexes in South Korea continues to rise, and consequently their electricity consumption is also increasing. This needs to be addressed as part of the nations efforts to reduce energy consumption. The objective of this research was to develop a decision support model for determining the need to improve multi-family housing complexes. In this research, 1664 cases located in Seoul were selected for model development. The research team collected the characteristics and electricity energy consumption data of these projects in 2009-2010. The following were carried out in this research: (i) using the Decision Tree, multi-family housing complexes were clustered based on their electricity energy consumption; (ii) using Case-Based Reasoning, similar cases were retrieved from the same cluster; and (iii) using a combination of Multiple Regression Analysis, Artificial Neural Network, and Genetic Algorithm, the prediction performance of the developed model was improved. The results of this research can be used as follows: (i) as basic research data for continuously managing several energy consumption data of multi-family housing complexes; (ii) as advanced research data for predicting energy consumption based on the project characteristics; (iii) as practical research data for selecting the most optimal multi-family housing complex with the most potential in terms of energy savings; and (iv) as consistent and objective criteria for incentives and penalties.

Comparison of Goldmann Applanation Tonometer, Noncontact Tonometer, and TonoPen XL for Intraocular Pressure Measurement in Different Types of Glaucomatous, Ocular Hypertensive, and Normal Eyes

PURPOSE To compare intraocular pressure (IOP) measurements by Goldmann applanation tonometer (GAT), noncontact tonometer (NCT), and TonoPen XL in glaucomatous, ocular hypertensive (OHT), and normal eyes, all with various central corneal thickness (CCT). METHODS In 508 subjects, IOP was measured by three different tonometers: GAT, NCT, and TonoPen. CCT was also measured. RESULTS There was a statistically significant difference between IOP measurements made using the three different tonometers (RMANOVA, P < 0.0001). Both NCT/GAT and TonoPen/GAT differences were significantly associated with age, CCT, IOP, and glaucoma diagnosis (multiple regression analysis, all P ≤ 0.05). CCT and IOP had the greatest effect on IOP measurement variance by different tonometers. IOP measurements made by NCT were consistently higher than those made by GAT; this difference increased both with IOP and CCT levels, but decreased with age and OHT. In addition, TonoPen overestimated the GAT in the eyes of subjects with OHT. CONCLUSION IOP measurement differences between the three tonometers are affected by age, type of glaucoma, and especially CCT and IOP levels.

Effect of signal strength on reproducibility of circumpapillary retinal nerve fiber layer thickness measurement and its classification by spectral-domain optical coherence tomography

PurposeTo assess the effect of signal strength (SS) on the reproducibility of circumpapillary retinal nerve fiber layer (cpRNFL) thickness measurement (measurement agreement) and its color-coded classification (classification agreement) by Cirrus spectral-domain optical coherence tomography (OCT).MethodsSeven hundred and eighty-nine participants, consecutively enrolled from December 2009 to August 2010, underwent two repeated Cirrus OCT scans to measure cpRNFL thickness (optic disc cube 200 × 200). Intraclass correlation (ICC) and linear-weighted kappa coefficient (κ) were calculated as indicators for cpRNFL measurement and classification agreement. The difference in cpRNFL thickness measurements between repeated OCT scans (inter-scan measurement difference) was correlated with mean SS, SS change or difference between repeated OCT scans (inter-scan SS difference), and average cpRNFL thickness.ResultsMean SS was negatively correlated with inter-scan SS difference and was positively correlated with the inter-scan measurement difference. Repeated scans with SS change (inter-scan SS difference = 1 or 2) showed larger variability than scans without SS change (inter-scan SS difference = 0) in total average cpRNFL and all quadrant maps except for the nasal quadrant. Multivariate analysis revealed that inter-scan measurement differences increased in subjects with thinner cpRNFL thicknesses and in scans with lower mean SS and higher inter-scan SS differences. Measurement agreement (ICC) in groups with larger inter-scan SS differences (=2) was lower than those with smaller inter-scan SS differences (=0 or 1) in their average cpRNFL and inferior quadrant maps. Classification agreement (κ) in groups with larger inter-scan SS differences (=2) was lower than those with smaller inter-scan SS differences (=0 or 1) in their average cpRNFL and quadrant maps (superior and inferior quadrant) although statistically not significant.ConclusionsSpecial attention should be paid when comparing the cpRNFL thickness measurement and color-coded classification of 2 or more Cirrus OCT scans as the agreements may be susceptible to SS differences.