The Power of AI

Introduction:

Artificial intelligence has been used to identify abnormal patterns in electrocardiograms and assist a clinician in making medial diagnosis. These systems have been trained to replicate a human understanding of the data and patterns. Neural networks have been developed to exceed the capabilities of board certified cardiologist in arrhythmia identification from single lead rhythm strips. (Cardiologist-Level Arrhythmia Detection with Convolutional Neural Networks)

In this study, I explore the idea of neural networks capability to extract meaningful information from a 12-lead ECG beyond what a human is capable of. Specifically, the use of a neural network to detect the gender of the patient from the conduction intervals and axis data from the ECG.

Methods

The conduction intervals and axis data was extracted from 93,740 deidentified 12 lead ECGs obtained through the Philips DXL ECG Algorithm in an automated process. This includes the following data points: RR interval, ie heart rate, PR interval, QRS duration, QT and QTc intervals as well as P wave axis, QRS axis and T wave axis. ECGs with missing values were excluded. Thus by this definition, patient in atrial fibrillation or flutter in which a PR interval could not be determined were excluded.  The age of the patient was manually entered by the technician obtaining the ECG tracing.

The data points were then normalized by taking the minimum and maximum values for each parameter and:

normalized x = (x-min) /(max-min)

The axis were also standardized to positive values. Thus an axis of -45 deg was encoded as 315 deg.

The gender was recorded as reported by the patient.