Most researchers construct their own methods by looking at relevant literature. A starting point may be to consider heart rate variability based on RR intervals. If possible, you may wish to segment your ECG into P waves, QRS, T, J waves and so on. Measurements are usually performed on these waves durations, amplitudes, and combination of.

I found your question interesting since signal processing methods (rather than image processing techniques) are typically applied to cardiac beat/rhythm classification problems.

Regarding the similarity of QRS complex profile and the "Mexican hat mother wavelet", I would suggest to either apply template matching with variations of this kernel or to apply 2D wavelet transform and extract relevant features.

Check this.

https://elcvia.cvc.uab.es/article/view/v18-n1-sunkara