Serial electrocardiography to detect newly emerging or aggravating cardiac pathology: a deep-learning approach

摘要: Serial electrocardiography aims to contribute electrocardiogram (ECG) diagnosis by comparing the ECG under consideration with a previously made in same individual. Here, we present novel algorithm construct dedicated deep-learning neural networks (NNs) that are specialized detecting newly emerging or aggravating existing cardiac pathology serial ECGs. We developed method for analysis and tested its performance detection of heart failure post-infarction patients, ischemia patients who underwent elective percutaneous coronary intervention. Core is repeated structuring learning procedure that, when fed 13 difference features (intra-individual differences in: QRS duration; QT interval; maximum; T-wave integral; complexity; ventricular gradient; QRS-T spatial angle; rate; J-point amplitude; symmetry), dynamically creates NN at most three hidden layers. An optimization process reduces possibility obtaining an inefficient due adverse initialization. Application our two clinical databases yielded 3-layer architectures, both showing high testing performances (areas receiver operating curves were 84% 83%, respectively). Our was successful two different clinical applications. Further studies will investigate if other problem-specific NNs can successfully be constructed, even it possible universal detect any pathologic change.