Optimization on Riemannian manifolds is an intuitive generalization of the traditional optimization algorithms in Euclidean spaces. In these algorithms, minimizing along a search direction becomes minimizing along a s...
详细信息
ISBN:
(纸本)9781467377058
Optimization on Riemannian manifolds is an intuitive generalization of the traditional optimization algorithms in Euclidean spaces. In these algorithms, minimizing along a search direction becomes minimizing along a search curve lying on a manifold. Computing such a curve to be subsequently searched upon is itself computational intensive. We propose a new minimization scheme aiming to find a better step size utilizing the first order information of the search curve. We prove that this scheme can provide further reduction for the cost function when the retraction and the vector transport are collinear. Then we adapt this scheme to propose a heuristic strategy for line search. In numerical experiments, we apply this heuristic strategy to one of the geometric algorithms for matrix completion and show its feasibility and the potential in accelerating computation.
Several ECG features are common electrocardiographic markers for manual interpretation of early repolarization (ER) and acute pericarditis (PCARD), both confounders for acute myocardial infarction (AMI). We hypothesiz...
详细信息
ISBN:
(纸本)9781479943722
Several ECG features are common electrocardiographic markers for manual interpretation of early repolarization (ER) and acute pericarditis (PCARD), both confounders for acute myocardial infarction (AMI). We hypothesized these features could improve automated AMI detection in the presence of ER and PCARD. Method: The training set of ECGs included cardiologist reading of ER (n= 147), PCARD (n= 114), normal (n=239) and AMI (n=380). AMI was confirmed by reading infarct evolution in serial ECGs. The test set came from emergency department chest pain patients (n= 1806). The reference was discharge diagnosis of AMI. Positive ECGs (n=1023) were both STEMI and NSTEMI. ECGs not meeting STEMI criteria by algorithm were excluded from both the test and training sets leaving 430 and 581 ECGs respectively. Two logistic regression AMI classifiers were compared, one using traditional features, another using traditional plus additional features to help detect ER and PCARD. Additional features included J-waves, notches, slurs, PQ segment depression, ST-T concavity, spatial QRS-T angle, and T-wave PCA ratio. Results: As expected, the traditional ST-T features had the most discrimination power. However, the automatically-selected best features included T-wave PCA ratio and the mean anterior PQ segment depression. Total accuracy was higher for the additional feature classifier, 79% versus 70% . Conclusion: Additional ECG features aimed at ER and PCARD may improve automatic AMI classification when STEMI criteria are met.
In this paper, we present a novel algorithm to evaluate the quality of ECG recordings. Our algorithm is designed to help clinicians in rapid selection of good quality ECG segments from long recordings collected by an ...
详细信息
ISBN:
(纸本)9781479943722
In this paper, we present a novel algorithm to evaluate the quality of ECG recordings. Our algorithm is designed to help clinicians in rapid selection of good quality ECG segments from long recordings collected by an ECG monitoring device such as a 12-lead bedside monitor. With some adjustments, we used the Computing in Cardiology Challenge 2011 database in order to compare the performance of our algorithm to the published results. The challenge was aimed to develop near real-time algorithms in mobile phones and provide feedback on quality of the ECGs for interpretation to the users who are mostly laypersons with little knowledge of ECG interpretation. Our algorithm generates a noise score which is a combination of two parameters: a high-frequency noise measure which accounts for the muscle noise and other fast changing artifacts, and a baseline wander noise measure quantifying the low-frequency noise. The training dataset (set A) with reference quality assessments was used to determine an optimum threshold on the ROC curve for classification of acceptable and unacceptable segments. The algorithm was then evaluated on the test dataset (set B) with undisclosed annotations. Our method achieved maximum accuracy of 93.9% on the training dataset and an accuracy of 90.2% on the test dataset, placing itself among the top 10 performers who participated in the challenge.
Misconnection of ECG lead-wires can generate abnormal ECG and erroneous diagnosis. Existing methods for detecting lead-wire interchange were designed for ECG devices using conventional lead system. In this work we dev...
详细信息
Misconnection of ECG lead-wires can generate abnormal ECG and erroneous diagnosis. Existing methods for detecting lead-wire interchange were designed for ECG devices using conventional lead system. In this work we developed an automatic ECG cable interchange detection algorithm and compared the algorithm performance between conventional and Mason-Likar (ML) electrode placements. The algorithm was developed based on a decision tree classifier which uses beat morphology measurements that were obtained using Philips DXL ECG algorithm. The algorithm was evaluated for detecting limb cable interchanges on an independent database which included both conventional and ML ECG recordings for each subject (total 423 subjects). There was no statistically significant difference in terms of overall sensitivity and specificity. This morphology-based cable interchange detection algorithm showed similarly high performance for maintaining a low false positive rate for both lead systems. Therefore, in practice, the same algorithm may be used with either electrode placement without a need for a special configuration.
Selecting a small informative subset from a given dataset, also called column sampling, has drawn much attention in machine learning. For incorporating structured data information into column sampling, research effort...
Selecting a small informative subset from a given dataset, also called column sampling, has drawn much attention in machine learning. For incorporating structured data information into column sampling, research efforts were devoted to the cases where data points are fitted with clusters, simplices, or general convex hulls. This paper aims to study nonconvex hull learning which has rarely been investigated in the literature. In order to learn data-adaptive nonconvex hulls, a novel approach is proposed based on a graph-theoretic measure that leverages graph cycles to characterize the structural complexities of input data points. Employing this measure, we present a greedy algorithmic framework, dubbed Zeta Hulls, to perform structured column sampling. The process of pursuing a Zeta hull involves the computation of matrix inverse. To accelerate the matrix inversion computation and reduce its space complexity as well, we exploit a low-rank approximation to the graph adjacency matrix by using an efficient anchor graph technique. Extensive experimental results show that data representation learned by Zeta Hulls can achieve state-of-the-art accuracy in text and image classification tasks.
The recent advancement of pacemaker technology: low power, biventricular (biV) pacing and adaptive pacing rate, has brought challenges for detection of pacemaker pulses (PPs) in the surface ECG. False positive PPs and...
详细信息
ISBN:
(纸本)9781479943722
The recent advancement of pacemaker technology: low power, biventricular (biV) pacing and adaptive pacing rate, has brought challenges for detection of pacemaker pulses (PPs) in the surface ECG. False positive PPs and undetected and unresolved PPs may consequently have a detrimental impact on a diagnostic ECG algorithm's rhythm or morphology interpretations. We have developed an algorithm to strengthen an existing PP detection algorithm using vector information to reject false positive PPs and detect the existence of a second biV PP closely spaced in time. We collected ECGs from both biV and non-biV pacemakers for algorithm development and performance validation. After training on the development dataset, our algorithm showed a biV paced rhythm detection sensitivity of 94.3% with a detection specificity of 99.3%.
暂无评论