Not sure what information you are thinking about.

But noise removal is basically some form of low-pass filtering, i.e. the removal of higher frequencies from the signal.

You may have a filter that low-pass filter that produces an output sample based on the value of the n last input samples. You will get loss of details from the signal and the signal must be way stronger than the noise for it to be meaningful.

For a repetitive signal, you can take a signal containing n periods of data and cut into n pieces and average the first sample of the n periods (possibly taking into account the values of a number of samples before and after) to produce the first sample of the output and then step to the second sample of the n periods. Stacking n signals like this will basically make the signal become n times stronger while not amplifying the noise. This allows the extraction of a repetitive signal that is weaker than the noise. But it one of the n periods have an anomaly, that anomaly will be reduced n times in the produced output. Also, if the patient has quickly changing BPM, you will not be able to find n periods of similar time length.

If the goal is to just capture the BPM, i.e. the heart rate, then you can get away with noisy signals and large amounts of filtering. But in this case the goal is to look at the actual shape of the curves. That means that the system may not perform any filtering that affects the curve shapes. That means that the input noise must be small compared to the actual heart measurements.

If the electrodes always produces a signal (possibly after running through a fixed amplifier) that is close to full range of the ADC, then you may probably not allow more than 1% noise in the signal. I don't know what amount of noise an expert would allow in the signal before thinking that the noise affects the use of the system.

But if the electrodes will produce a signal that may have very varying amplitude, then you will either have to use a preamplifier with selectable gain, or you will have to use an ADC with extra bits and then perform a software Automatic Gain Control (AGC) to scale the ADC readings to maybe consume 80% of the full-range scale of the display, with the remaining 20% reserved for spikes on the curve. In that case, the noise requirements will scale, i.e. if the ADC will sometimes only get pulses using 25% of the ADC range, then the noise level will have to be limited to about 1% of these 25% - the total noise level must then be about 0.25% for full-range data. That is around +/- 1 bit for a 10-bit ADC.

As I have said before, you should not focus on noise reduction on the sampled data. You should focus on minimizing the noise instroduced in the hardware. The sampled curves are not a cyclic repetition of identical heart beats, and a big goal is to catch noise produced by the heart while not being fooled by noise introduced by the hardware or noise removed by the software.