In fact, vital sign monitoring has become part of our daily lives, emerging as a trend of smart wearable devices used to manage health. Cardiac arrhythmia is defined as any variation in the normal rhythm of the heartbeat, such that the heart beats inappropriately. This work presents a study on the classification of ECG signals into 4 classes: Normal (N), Supraventricular Ectopic (SVE), Ventricular Ectopic (VE), and Fusion of Normal and Ventricular (F). Using the MIT-BIH arrhythmia database and Python programming through the Keras and TensorFlow libraries, it was possible to investigate the performance of the Resnet50 with raw images of ECG signal, with the continuous wavelet transform spectrum, and with the concatenation of the raw image and the continuous wavelet transform spectrum, that is, the features extracted from the images through two Resnet50 networks, with very close performances, ranging from 93% of accuracy, precision, and recall for the raw signal and the concatenation, and 94% of accuracy, precision and recall for the continuous wavelet transform spectrum.