Zhi Xiong Koh

Landmark recognition with sparse representation classification and extreme learning machine

Abstract Along with the rapid development of intelligent mobile terminals, applications on landmark recognition attract increasingly attentions by world wide researchers in the past several years. Although promising achievements have been presented, designing a robust recognition system with an accurate recognition rate and fast response speed is still challenging. To address these issues, we propose a novel landmark recognition algorithm in this paper using the spatial pyramid kernel based bag-of-words (SPK-BoW) histogram approach with the feedforward artificial neural networks (FNN) and the sparse representation classifier (SRC). In the proposed algorithm, the SPK-BoW approach is first employed to extract features and construct an overcomplete dictionary for landmark image representation. Then, the FNN trained with the extreme learning machine (ELM) algorithm combined with the SRC is implemented for landmark image recognition. We conduct experiments using the Nanyang Technological University (NTU) campus landmark database to show that the proposed method achieves a high recognition rate than ELM and a lower response time than the sparse representation technique.

Prediction of cardiac arrest in critically ill patients presenting to the emergency department using a machine learning score incorporating heart rate variability compared with the modified early warning score.

IntroductionA key aim of triage is to identify those with high risk of cardiac arrest, as they require intensive monitoring, resuscitation facilities, and early intervention. We aim to validate a novel machine learning (ML) score incorporating heart rate variability (HRV) for triage of critically ill patients presenting to the emergency department by comparing the area under the curve, sensitivity and specificity with the modified early warning score (MEWS).MethodsWe conducted a prospective observational study of critically ill patients (Patient Acuity Category Scale 1 and 2) in an emergency department of a tertiary hospital. At presentation, HRV parameters generated from a 5-minute electrocardiogram recording are incorporated with age and vital signs to generate the ML score for each patient. The patients are then followed up for outcomes of cardiac arrest or death.ResultsFrom June 2006 to June 2008 we enrolled 925 patients. The area under the receiver operating characteristic curve (AUROC) for ML scores in predicting cardiac arrest within 72 hours is 0.781, compared with 0.680 for MEWS (difference in AUROC: 0.101, 95% confidence interval: 0.006 to 0.197). As for in-hospital death, the area under the curve for ML score is 0.741, compared with 0.693 for MEWS (difference in AUROC: 0.048, 95% confidence interval: -0.023 to 0.119). A cutoff ML score ≥ 60 predicted cardiac arrest with a sensitivity of 84.1%, specificity of 72.3% and negative predictive value of 98.8%. A cutoff MEWS ≥ 3 predicted cardiac arrest with a sensitivity of 74.4%, specificity of 54.2% and negative predictive value of 97.8%.ConclusionWe found ML scores to be more accurate than the MEWS in predicting cardiac arrest within 72 hours. There is potential to develop bedside devices for risk stratification based on cardiac arrest prediction.

Heart rate variability risk score for prediction of acute cardiac complications in ED patients with chest pain

BACKGROUND We aimed to develop a risk score incorporating heart rate variability (HRV) and traditional vital signs for the prediction of early mortality and complications in patients during the initial presentation to the emergency department (ED) with chest pain. METHODS We conducted a prospective observational study of patients with a primary complaint of chest pain at the ED of a tertiary hospital. The primary outcome was a composite of mortality, cardiac arrest, ventricular tachycardia, hypotension requiring inotropes or intraaortic balloon pump insertion, intubation or mechanical ventilation, complete heart block, bradycardia requiring pacing, and recurrent ischemia requiring revascularization, all within 72 hours of arrival at ED. RESULTS Three hundred nine patients were recruited, and 25 patients met the primary outcome. Backwards stepwise logistic regression was used to derive a scoring model that included heart rate, systolic blood pressure, respiratory rate, and low frequency to high frequency ratio. For predicting complications within 72 hours, the risk score performed with an area under the curve of 0.835 (95% confidence interval [CI], 0.749-0.920); and a cutoff of 4 and higher in the risk score gave a sensitivity of 0.880 (95% CI, 0.677-0.968), specificity of 0.680 (95% CI, 0.621-0.733), positive predictive value of 0.195, and negative predictive value of 0.985. The risk score performed better than ST elevation/depression and troponin T in predicting complications within 72 hours. CONCLUSION A risk score incorporating heart rate variability and vital signs performed well in predicting mortality and other complications within 72 hours after arrival at ED in patients with chest pain.

Improved neurologically intact survival with the use of an automated, load-distributing band chest compression device for cardiac arrest presenting to the emergency department.

IntroductionIt has been unclear if mechanical cardiopulmonary resuscitation (CPR) is a viable alternative to manual CPR. We aimed to compare resuscitation outcomes before and after switching from manual CPR to load-distributing band (LDB) CPR in a multi-center emergency department (ED) trial.MethodsWe conducted a phased, prospective cohort evaluation with intention-to-treat analysis of adults with non-traumatic cardiac arrest. At these two urban EDs, systems were changed from manual CPR to LDB-CPR. Primary outcome was survival to hospital discharge, with secondary outcome measures of return of spontaneous circulation, survival to hospital admission and neurological outcome at discharge.ResultsA total of 1,011 patients were included in the study, with 459 in the manual CPR phase (January 01, 2004, to August 24, 2007) and 552 patients in the LDB-CPR phase (August 16, 2007, to December 31, 2009). In the LDB phase, the LDB device was applied in 454 patients (82.3%). Patients in the manual CPR and LDB-CPR phases were comparable for mean age, gender and ethnicity. The mean duration from collapse to arrival at ED (min) for manual CPR and LDB-CPR phases was 34:03 (SD16:59) and 33:18 (SD14:57) respectively. The rate of survival to hospital discharge tended to be higher in the LDB-CPR phase (LDB 3.3% vs Manual 1.3%; adjusted OR, 1.42; 95% CI, 0.47, 4.29). There were more survivors in LDB group with cerebral performance category 1 (good) (Manual 1 vs LDB 12, P = 0.01). Overall performance category 1 (good) was Manual 1 vs LDB 10, P = 0.06.ConclusionsA resuscitation strategy using LDB-CPR in an ED environment was associated with improved neurologically intact survival on discharge in adults with prolonged, non-traumatic cardiac arrest.

EZ-IO in the ED: an observational, prospective study comparing flow rates with proximal and distal tibia intraosseous access in adults ☆

INTRODUCTION Intraosseous (IO) access is an important alternative to conventional intravenous access when intravenous access is difficult. METHODS A nonrandomized, prospective, observational study comparing flow rates with distal and proximal tibia IO access in adults using the EZ-IO-powered drill device. The proximal tibia was the first site of insertion, and a second IO was inserted in the distal tibia if clinically indicated. Intravenous saline infusion was started for all patients, initially without, then with a pressure bag device applied. RESULTS From September 19, 2008 to November 3, 2010, 22 patients were recruited, with 20 proximal tibial and 22 distal tibia insertions. Two patients had only distal tibia IO insertions. Five distal tibia and 3 proximal tibia insertions had no flow when initiating normal saline infusion without pressure. Upon comparing the mean flow rates without pressure bag, it is significantly faster in the proximal tibia, 4.96 mL/min, compared with distal tibia, 2.07 ml/min, difference of 2.89 ml/min (95% CI 1.20-4.58). Flow rates with pressure bags also revealed a similar result. Flow rates in the proximal tibia were significantly faster, 7.70 ml/min to that of distal tibia, 3.80 ml/min, difference of 3.89 ml/min (95% CI 1.68-6.10). In both proximal and distal tibia groups, the flow rates are also significantly faster with pressure bags compared with without. CONCLUSION Flow rates are significantly faster in the proximal tibia compared with the distal tibia. In addition, flow rates with pressure bags are significantly faster than without pressure bags in both groups.

Risk Scoring for Prediction of Acute Cardiac Complications from Imbalanced Clinical Data

Fast and accurate risk stratification is essential in the emergency department (ED) as it allows clinicians to identify chest pain patients who are at high risk of cardiac complications and require intensive monitoring and early intervention. In this paper, we present a novel intelligent scoring system using heart rate variability, 12-lead electrocardiogram (ECG), and vital signs where a hybrid sampling-based ensemble learning strategy is proposed to handle data imbalance. The experiments were conducted on a dataset consisting of 564 chest pain patients recruited at the ED of a tertiary hospital. The proposed ensemble-based scoring system was compared with established scoring methods such as the modified early warning score and the thrombolysis in myocardial infarction score, and showed its effectiveness in predicting acute cardiac complications within 72 h in terms of the receiver operation characteristic analysis.

Comparing HEART, TIMI, and GRACE scores for prediction of 30-day major adverse cardiac events in high acuity chest pain patients in the emergency department

BACKGROUND The HEART, TIMI, and GRACE scores have been applied in the Emergency Department (ED) to risk stratify patients with undifferentiated chest pain. This study aims to compare the accuracy of HEART, TIMI, and GRACE for the prediction of major adverse cardiac events (MACE) in high acuity chest pain patients. METHODS Adult patients who presented with chest pain suggestive of cardiac origin in the most acute triage category at an academic ED from September 2010 to October 2015 were included. The HEART, TIMI, and GRACE scores were calculated retrospectively from prospectively collected data. The primary outcome was occurrence of MACE (mortality, AMI, PCI, CABG) within 30-days of initial presentation. RESULTS 604 patients were included in the study. Patient demographics include an average age of 61years, 69% male, and 48% with history of ischemic heart disease. 36% of patients met the primary outcome. The c-statistics of HEART, TIMI, and GRACE were 0.78 (95% CI: 0.74-0.81), 0.65 (95% CI: 0.60-0.69), and 0.62 (95% CI: 0.58-0.67), respectively. For the purpose of accurately ruling out patients for 30-day MACE, a HEART score of ≤3 had a sensitivity and NPV of 99% and 98%, respectively, compared to 97% and 91%, respectively, for TIMI=0, and 94% and 85%, respectively, for GRACE ≤75. The percent of patients with 30-day MACE with HEART scores between 0 and 3, 4-6, and 7-10 was 2%, 28%, and 63%, respectively. CONCLUSION In high acuity chest pain patients, the HEART score is superior to the TIMI and GRACE scores in predicting 30-day MACE.

Risk stratification for prediction of adverse coronary events in emergency department chest pain patients with a machine learning score compared with the TIMI score

a Department of Emergency Medicine, Singapore General Hospital, Singapore b Centre for Quantitative Medicine, Duke-NUS Graduate Medical School, Singapore c SingHealth Emergency Medicine Residency Program, Singapore Health Services, Singapore d Department of Statistics and Applied Probability, National University of Singapore, Singapore e Division of Research, Singapore General Hospital, Singapore f Singapore Institute of Technology, Singapore g School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore h Health Services and Systems Research, Duke-NUS Graduate Medical School, Singapore

Evolutionary Voting-Based Extreme Learning Machines

Voting-based extreme learning machine (V-ELM) was proposed to improve learning efficiency where majority voting was employed. V-ELM assumes that all individual classifiers contribute equally to the decision ensemble. However, in many real-world scenarios, this assumption does not work well. In this paper, we aim to enhance V-ELM by introducing weights to distinguish the importance of each individual ELM classifier in decision making. Genetic algorithm is used for optimizing these weights. This evolutionary V-ELM is named as EV-ELM. Results on several benchmark databases show that EV-ELM achieves the highest classification accuracy compared with V-ELM and ELM.

Ensemble-Based Risk Scoring with Extreme Learning Machine for Prediction of Adverse Cardiac Events

Accurate prediction of adverse cardiac events for the emergency department (ED) chest pain patients is essential in risk stratification due to the current ambiguity in diagnosing acute coronary syndrome. While most current practices rely on human decision by measuring clinical vital signs, computerized solutions are gaining popularity. We have previously proposed an ensemble-based scoring system (ESS). In this paper, we aim to extend the ESS system using extreme learning machine (ELM), a fast learning algorithm for neural networks. We recruited patients from the ED of Singapore General Hospital, and extracted features such as heart rate variability, 12-lead ECG parameters, and vital signs. We also proposed a novel algorithm called ESS-ELM to predict adverse cardiac events. Different from the original ESS algorithm, ESS-ELM uses the under-sampling technique only in model training. Our proposed method was compared to the original ESS algorithm and several clinical scores in predicting patient outcome. With a cohort of 797 recruited patients, we demonstrated that ESS-ELM outperformed the original ESS algorithm and three established clinical scores, namely HEART, TIMI, and GRACE, in terms of receiver operating characteristic analysis. Furthermore, we have investigated the impact of hidden node number and ensemble size on the predictive performance. ELM has demonstrated the flexibility in its integration with the ESS algorithm. Experiments showed the value of ESS-ELM in prediction of adverse cardiac events. Future works may include the use of new ELM-based learning methods and further validation with a new cohort of patients.