Hyunjin Lee

Clinical outcomes of exercise-induced pulmonary hypertension in subjects with preserved left ventricular ejection fraction: implication of an increase in left ventricular filling pressure during exercise

Objective To investigate clinical outcomes of exercise-induced pulmonary hypertension (PH) and implications of an increase in left ventricular (LV) filling pressure during exercise in subjects with preserved LV ejection fraction. Design Longitudinal follow-up study. Setting Subjects who were referred for diastolic stress echocardiography. Patients and methods The ratio of transmitral and annular velocities (E/Ea) and pulmonary artery systolic pressure (PASP) at rest and during exercise were measured in 498 subjects (57±11 years; 201 male). Exercise-induced PH was defined as present if PASP ≥50 mm Hg at 50 W of exercise, and an increase in LV filling pressure during exercise was present if E/Ea ≥15 at 50 W. Main outcome measures A combination of major cardiovascular events and any cause of death. Results During a median follow-up of 41 months, there were 14 hospitalisations and four deaths. Subjects with exercise-induced PH had significantly worse clinical outcomes than those without (p=0.014). Subjects with exercise-induced PH associated with an increase in E/Ea during exercise had significantly worse outcomes than other groups (p<0.001). However, prognosis was similar between subjects with exercise-induced PH without an increase in E/Ea and those without exercise-induced PH. In subjects with exercise-induced PH, E/Ea at 50 W was an independent predictor of adverse outcomes (HR 1.37; 95% CI 1.02 to 1.83; p=0.036). Conclusions Exercise-induced PH provides prognostic information in subjects with preserved LV ejection fraction. The excess risk of exercise-induced PH is restricted to subjects with an increase in estimated LV filling pressure during exercise.

Impact of left ventricular longitudinal diastolic functional reserve on clinical outcome in patients with type 2 diabetes mellitus

Background Left ventricular longitudinal diastolic functional reserve (DFR), as assessed by the change in early diastolic mitral annular velocity (E′) during exercise, is abnormal in patients with type 2 diabetes mellitus (DM). However, the impact of left ventricular longitudinal DFR on clinical outcome has not been explored. This study evaluated the incremental prognostic value of left ventricular DFR in patients with type 2 DM without overt heart disease. Methods Of 1485 patients who were referred for exercise stress echocardiography, 197 consecutive patients (mean age, 58 years; 84 men) with type 2 DM without overt heart disease were identified. Left ventricular longitudinal DFR was defined as the change in E′ from resting to exercise (ΔE′). The endpoint was a composite of death and hospitalisation for heart failure (HF). Results During a median follow-up of 57 months (range 6–90), 18 of 197 patients (9.1%) had adverse events (12 deaths, six hospitalisations for HF). Independent predictors of adverse events in a Cox regression analysis were estimated glomerular filtration rate (HR 0.97; 95% CI 0.95 to 0.98; p<0.001), DM duration (HR 1.07; 95% CI 1.01 to 1.14; p=0.018) and ΔE′ (HR 0.58; 95% CI 0.40 to 0.85; p=0.005). In an incremental model, the addition of stress echo data significantly increased the χ2 of the clinical and resting left ventricular function model, from 40.5 to 46.6 (p=0.005). Conclusion Assessment of left ventricular longitudinal DFR during exercise provided incremental prognostic information in patients with type 2 DM without overt heart disease.

Visualization Method of Document Retrieval Result based on Centers of Clusters

Because it is difficult on existing document retrieval systems to visualize the search result, search results show document titles and short summaries of the parts that include the search keywords. If the result list is long, it is difficult to examine all the documents at once and to find a relation among them. This study uses clustering to classify similar documents into groups to make it easy to grasp the relations among the searched documents. Also, this study proposes a two-level visualization algorithm such that, first, the center of clusters is projected to low-dimensional space by using multi-dimensional scaling to help searchers grasp the relation among clusters at a glance, and second, individual documents are drawn in low-dimensional space based on the center of clusters using the orbital model as a basis to easily confirm similarities among individual documents. This study is tested on the benchmark data and the real data, and it shows that it is possible to visualize search results in real time.

Network Optimization through Learning and Pruning in Neuromanifold

In this paper, we propose an optimization method of neural networks based on the geometrical structure of neuromanifold. The optimizing process starts from the manifold of sufficiently large network model. In the manifold of the given network structure, we first find an optimal point, which achieves good generalization performance. To do this, we propose an extension of the adaptive natural gradient learning with regularization term. Using hierarchical structure of neuromanifold, we then try to optimize the network structure. To do this, we apply the natural pruning method starting from the current optimal parameter point. The whole optimization process can be explained from the geometrical point of view. We confirm the generalization performance of the optimized network by the proposed method through experiments on benchmark data sets.

Visualization of Document Retrieval using External Cluster Relationship

Owing to the limitations of existing visualization schemes, existing document retrieval systems display limited results, often showing only document titles, short summaries, and keywords. This makes it difficult to examine multiple results at once or to find a meaningful relationship between results. This study proposes a new method for the real-time visualization of document retrieval results via clustering. The method clusters similar documents into groups, making it easier to understand the relationship between the retrieved documents. This study also proposes a two-level visualization algorithm which projects the cluster centers onto a two-dimensional space using multidimensional scaling in order to illustrate the relationships among different clusters, and displays individual documents at locations determined by the external cluster relationship in low dimensional space in order to allow the comparison of individual documents. The method was tested on benchmark data and real-world data, and the results show that it is possible to visualize the search results in real time.

Fast K-Means Clustering Algorithm using Prediction Data

In this paper we proposed a fast method for a K-Means Clustering algorithm. The main characteristic of this method is that it uses precalculated data which possibility of change is high in order to speed up the algorithm. When calculating distance to cluster centre at each stage to assign nearest prototype in the clustering algorithm, it could reduce overall computation time by selecting only those data with possibility of change in cluster is high. Calculation time is reduced by using the distance information produced by K-Means algorithm when computing expected input data whose cluster may change, and by using such distance information the algorithm could be less affected by the number of dimensions. The proposed method was compared with original K-Means method - Lloyd`s and the improved method KMHybrid. We show that our proposed method significantly outperforms in computation speed than Lloyd`s and KMHybrid when using large size data which has large amount of data, great many dimensions and large number of clusters.

Determining the number of Clusters in On-Line Document Clustering Algorithm

Clustering is to divide given data and automatically find out the hidden meanings in the data. It analyzes data, which are difficult for people to check in detail, and then, makes several clusters consisting of data with similar characteristics. On-Line Document Clustering System, which makes a group of similar documents by use of results of the search engine, is aimed to increase the convenience of information retrieval area. Document clustering is automatically done without human interference, and the number of clusters, which affect the result of clustering, should be decided automatically too. Also, the one of the characteristics of an on-line system is guarantying fast response time. This paper proposed a method of determining the number of clusters automatically by geometrical information. The proposed method composed of two stages. In the first stage, centers of clusters are projected on the low-dimensional plane, and in the second stage, clusters are combined by use of distance of centers of clusters in the low-dimensional plane. As a result of experimenting this method with real data, it was found that clustering performance became better and the response time is suitable to on-line circumstance.

Reconsideration to pruning and regularization for complexity optimization in neural networks

The ultimate purpose of neural network design is to find an optimal network that can give good generalization performance with compact structure. To achieve this, it is necessary to control complexities of networks so as to avoid its overfitting to noisy learning data. The most popular methods for complexity control are the pruning method and the regularization method. Even though there have been many variations in the methods, the peculiar properties of each method compared to others has not been so clear. We reconsider the pruning strategy from a geometrical and statistical viewpoint, and show that the natural pruning method is in accordance with the geometrical and statistical intuition in choosing connections to be pruned. In addition, we also suggest that the regularization method should be used in combination with natural pruning in order to improve the optimization performance. We also show some experimental results supporting our suggestions.