In areas like telecommunications, you might not know the channel characteristics, you know the source signal and received signal. With biomedical signals you have no access to the source signal. As noted the channel characteristics are unknown. There is often a lot of noise and cross-talk, which overlap in time and frequency (and you can't often "turn off" the other physiological sources or block them.

Analysis becomes difficult because there is lack of ground truth of the actual signal you are trying to work with.

Biological systems are also highly interconnected subsystems and exhibit characteristics that are nonlinear, stochastic, deterministic, and nonstationary. This signal, noise, and measurement is highly variable and there is a low SNR to contend with as well.

Problems like

- variable contact resistance of the skin; for EMG the subject has to move in some researches, making acquisition more difficult

- hum can be 60dB over useful signal in 50Hz component, electrical network can carry another components than usual 50Hz and harmonics of 50Hz

- complexity of biological tissue, which makes modelling very hard

- difficulties linked to a standard reference generator for such signals

- dependence of the signals of many unknown factors (bigger the tissue in the current path, more unknown factors), dependence of the signal of emotional state of the patient (for EEG can be an advantage, but emotional states are not measurable in numbers anyway)

- sometimes invasive acquisition of the signals (for EEG you have to remove hair for instance)

- if you want to localize the source of potential, huge number of electrodes required

All those factors mentioned by Teodor result in a high variability in the patient or test-animal group.

I agree with all the comments. But these are the sources of signals. Once we have the signals, what about analysis? All real time signals are random in nature, so how analysis is different.

Yes, variable contact resistance of the skin, and very weak (low voltage) signals are the problems, but these are the issues with signal acquisition.

The signal analysis is not more difficult, but -because of the variability- one has to be careful with drawing conclusions

Most of the times Pattern Recognition in Biomedical Signal (like EEG) is difficult.

In areas like telecommunications, you might not know the channel characteristics, you know the source signal and received signal. With biomedical signals you have no access to the source signal. As noted the channel characteristics are unknown. There is often a lot of noise and cross-talk, which overlap in time and frequency (and you can't often "turn off" the other physiological sources or block them.

Analysis becomes difficult because there is lack of ground truth of the actual signal you are trying to work with.

Biological systems are also highly interconnected subsystems and exhibit characteristics that are nonlinear, stochastic, deterministic, and nonstationary. This signal, noise, and measurement is highly variable and there is a low SNR to contend with as well.

That depends on what kind of bio-medical signals are you acquiring. For example, in case of EEG, noise is a huge issue. In fNIRS signals scattering and optical path-length estimation are difficult, etc.

Biomedical signal analysis is challenging but still in the same order of difficulties as other physical signals. Source localization, modeling and acquisition of such signals represent the problematic part of biomedical signal processing.

I would say, BEING ABLE TO FIND SAMPLES and the REPEATABILITY of RESULTS. I will only speak in regards to ECG signals, since I am collaborating with somebody that is an expert in it, and this is our research area. Due to the FDA (and other) government regulations, you can't simply go get sample data from a patient any time you want. There are major privacy laws etc ... So, this significantly limits your ability to find repeatable samples. Here, in our university, we have a database called THEW for Holter ECG signals (a worldwide free-to-access database). This is addressing the problem of finding sample data a lot. But, otherwise, getting the hospital to apply even the simplest tests to even a VOLUNTEER patient is major paperwork (if not outright impossible).

This means that, not being able to find consistent samples for a researcher eventually makes the analysis of the bio signals difficult, since the samples are working with are inconsistent at best. This also makes your results difficult to REPEAT. Bio signals are very susceptible to an individual's overall health conditions, and repeating the same results from one researcher to another seems very difficult, even for extremely well structured (well defined) signals like ECG.

Nahed, this is a very important point ! Good thing you brought it up ! I agree ! The noise that is ambient is always the same amplitude (say, in mV or uV levels, depending on the environment). So, unless you can put somebody in a Faraday cage and acquire the bio signals (doesn't sound practical !), you are prone to have a worse SNR (signal-to-noise ratio) for bio signals , since the levels (say, ECG) are in the 10-20 mV range ...

Two factors that immediately spring to mind:

1. Signal-to-noise ratio;

2, Difficult (or unethical) to stimulate to measure response as we cannot apply stimulus (kick it and see how it responds). This is particularly true for large stimuli.

@Tolga Soyata... the ECG is NOT in the 10-20 mV range.

Hi Fernando, quick question to clarify a few points:

1) The "published" rates on ECG signal levels are 40 uV - 3 mV. I agree, 10-20mV is way too high and would never reach these levels, even for widely varying levels from one patient to another. What is your typical experience with the signal levels ?

2) The typical setup I use is the TI DSP board with LL, RL, LA, RA electrodes. I do not have too much experience with 12-lead "practical" hardware. We only use synthetic data from the database. Do the measurements make a big difference with 12-leads ?

3) Do you have any practical numbers on what the NOISE LEVELS should be ? which will lead the discussion to SNR ...

@Tolga Soyata

1. Yes;

2. Tre precordials allow the doctors to 'see' the 3rd dimension to the ECG vector;

3. The higher the better.

In addition to the fact that Biological systems are highly interconnected subsystems and exhibit characteristics that are nonlinear, stochastic, non-deterministic, and non-stationary, there is lot of subjectivity in the biological systems i.e. there is lot of variation from case to case in the characteristics of even the same type of signal.

As Nahed and Tolga pointed out the weak signal characteristic is very important. Weak signal coupled with intra-subject variability (e.g. anxiety level to the test itself ) and inter test subject variability make it more challenging in drawing a baseline. Other challenges is for example when you filter a signal you may be loosing specific information that may be relevant especially when there is so much variability (especially if you do not have a well established baseline).

While obtaining a baseline is possible is has to be done statistically but this implies it will be very expensive and therefore it may be difficult to reproduce.

Biomedical signal analysis is concerned with the problem of studying physiological systems generating those signals. As is well known, physiological system are time-varying, non-stationary, non-linear and this fact makes the application of analysis tools (USUALLY made for linear, stationary systems analysis) to the study of biomedical signals very difficult. Moreover, their amplitudes are very small compared with the enviromental noise. Hard, very-very hard job. For this reason there exist the Biomedical Engineer (such as me!)

Bio medical signals are Non linear and random in nature. Bio medical signals has frequency contents from 0.05 Hz. Content of Bio medical signal is closely related to interference frequency also.