Diagnostic Model of in-Hospital Mortality in Patients with Acute ST-Segment Elevation Myocardial Infarction Used Artificial Intelligence Methods : Algorithm Development and Validation (Preprint)

BACKGROUND Preventing in-hospital mortality in Patients with ST-segment elevation myocardial infarction (STEMI) is a crucial step. OBJECTIVE The objective of our research was to to develop and externally validate the diagnostic model of in-hospital mortality in acute STEMI patients used artificial intelligence methods. METHODS As our datasets were highly imbalanced, we evaluated the effect of down-sampling methods. Therefore, down-sampling techniques was additionally implemented on the original dataset to create 1 balanced datasets. This ultimately yielded 2 datasets; original, and down-sampling. We divide non-randomly the American population into a training set and a test set , and anther American population as the validation set. We used artificial intelligence methods to develop and externally validate the diagnostic model of in-hospital mortality in acute STEMI patients, including logistic regression, decision tree, extreme gradient boosting (XGBoost), K nearest neighbor classification model ,and multi-layer perceptron.We used confusion matrix combined with the area under the receiver operating characteristic curve (AUC) to evaluate the pros and cons of the above models. RESULTS The strongest predictors of in-hospital mortality were age, female, cardiogenic shock, atrial fibrillation(AF), ventricular fibrillation(VF),in-hospital bleeding and medical history such as hypertension, old myocardial infarction.The F2 score of logistic regression in the training set, the test set , and the validation data set were 0.7, 0.7, and 0.54 respectively.The F2 score of XGBoost were 0.74, 0.52, and 0.54 respectively. The F2 score of decision tree were 0.72, 0.51,and 0.52 respectively. The F2 score of K nearest neighbor classification model were 0.64,0.47, and 0.49 respectively. The F2 score of multi-layer perceptron were 0.71, 0.54, and 0.54 respectively. The AUC of logistic regression in the training set, the test set, and the validation data set were 0.72, 0.73, and 0.76 respectively. The AUC of XGoBost were 0.75, 0.73, and 0.75 respectively. The AUC of decision tree were 0.75, 0.71,and 0.74 respectively. The AUC of K nearest neighbor classification model were 0.71,0.69, and 0.72 respectively. The AUC of multi-layer perceptron were 0.73, 0.74, and 0.75 respectively. The diagnostic model built by logistic regression was the best. CONCLUSIONS The strongest predictors of in-hospital mortality were age, female, cardiogenic shock, AF, VF,in-hospital bleeding and medical history such as hypertension, old myocardial infarction. We had used artificial intelligence methods developed and externally validated the diagnostic model of in-hospital mortality in acute STEMI patients.The diagnostic model built by logistic regression was the best. CLINICALTRIAL We registered this study with WHO International Clinical Trials Registry Platform (ICTRP) (registration number: ChiCTR1900027129; registered date: 1 November 2019). http://www.chictr.org.cn/edit.aspx?pid=44888&htm=4.

Improving the Accuracy of Nearest-Neighbor Classification Using Principled Construction and Stochastic Sampling of Training-Set Centroids

A conceptually simple way to classify images is to directly compare test-set data and training-set data. The accuracy of this approach is limited by the method of comparison used, and by the extent to which the training-set data cover configuration space. Here we show that this coverage can be substantially increased using coarse-graining (replacing groups of images by their centroids) and stochastic sampling (using distinct sets of centroids in combination). We use the MNIST and Fashion-MNIST data sets to show that a principled coarse-graining algorithm can convert training images into fewer image centroids without loss of accuracy of classification of test-set images by nearest-neighbor classification. Distinct batches of centroids can be used in combination as a means of stochastically sampling configuration space, and can classify test-set data more accurately than can the unaltered training set. On the MNIST and Fashion-MNIST data sets this approach converts nearest-neighbor classification from a mid-ranking- to an upper-ranking member of the set of classical machine-learning techniques.

Acceleration of Large Margin Metric Learning for Nearest Neighbor Classification Using Triplet Mining and Stratified Sampling

Metric learning is a technique in manifold learning to find a projection subspace for increasing and decreasing the inter- and intra-class variances, respectively. Some metric learning methods are based on triplet learning with anchor-positive-negative triplets. Large margin metric learning for nearest neighbor classification is one of the fundamental methods to do this. Recently, Siamese networks have been introduced with the triplet loss. Many triplet mining methods have been developed for Siamese nets; however, these techniques have not been applied on the triplets of large margin metric learning. In this work, inspired by the mining methods for Siamese nets, we propose several triplet mining techniques for large margin metric learning. Moreover, a hierarchical approach is proposed, for acceleration and scalability of optimization, where triplets are selected by stratified sampling in hierarchical hyper-spheres. We analyze the proposed methods on three publicly available datasets.