The importance of disease classification studies for the development of diagnosis & prognosis systems is crucial. Here the main problem is the lack of available medical datasets to use.
For EMG diagnosis, many studies used EMGLab datasets, but the website is NOT available currently.
https://www.emglab.net/emglab/Signals/N2001/N2001.php?diagnosis=ALS
https://www.emglab.net/emglab/Signals/
https://www.emglab.net/emglab/Signals/signals.php
And I can't find any open-access EMG diagnosis dataset. Parkinson's, SMA, ALS, or other. There are only hand gesture detection datasets present.
Other options are
1- Using an EMG simulation program like EMG-GAN (I don't know how efficient and representative it is)
2- Long-term clinical study with a budget
3- Somehow finding a -retrospective- signal record from a hospital (I don't know how to reach them)
Additionally, I want to know the usability of this short-term prototyping approach :
Simulation of signals by mimicking the patient movements, captured by sEMG device purchased. Can this, highly questionable and insufficient method, be used for the first analysis?
I want to hear your suggestions, thoughts, and shares about the topic :)
Pinheiro, W.C.; Bittencourt, B.E.; Luiz, L.B.; Marcello, L.A.; Antonio, V.F.; de Lira, P.H.A.; Stolf, R.G.; Castro, M.C.F. Parkinson’s Disease Tremor Suppression. In Proceedings of the 10th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2017), Porto, Portugal, 21–23 February 2017; pp. 149–155
The authors of this paper have a pathological EMG dataset. It is private, but if you request the authors that you need it for your research, they may accede to your request.
Oğuzhan Memiş Wa Alaikum Salaam, you raise an excellent point - access to high quality, representative EMG datasets is a major bottleneck, especially for pathological signals. Unfortunately EMGLab shutting down made things much harder. Here are some suggestions that may help:
- NIH funded an NIH-EMG dataset for gesture recognition that also includes some pathology signals like dystonia, myopathy. It requires a license agreement but is freely available for research.
- Recent papers like Peng et al 2022 in IEEE JBHI released multi-class pathological EMG data for public benchmarking. Reach out to authors directly.
- PhysioNet has some older but still useful EMG data for conditions like dystonia and neuropathy. Limited but standardized.
- Rather than simulations, you may have more luck partnering with a local hospital to record (with ethics approval) from a small cohort of patients. Get multi-day data.
- Evaluating robustness of methods on highly multimodal synthetic data is still valuable initially. Just be cautious about clinical claims without real validation.
- Surveys and competitions inviting clinical submissions have had some success aggregating data from dispersed sources. Explore collaborative options.
I know the data access challenges are frustrating, but there are still paths forward.
Qamar Ul Islam Salam Alaikum again, thank you for your kind answers and suggestions, and I apologize for my late return.
1) I don't see any pathology information in the NIH-EMG article, what was the name of the project?
2) I can't find that multi-class pathological EMG data and related article from Peng in JBHI journal, can you share the article name? When I looked at JBHI I found ECG study but it is not what I am looking for. And of course there were various EMG gesture articles.
3) I can not see any pathological EMG dataset in PhysioNet, there are only 3 signal examples from myopathy-health-neuropathy.
4) What you suggest about partnering a local hospital , how to reach them and retrospective data sharing ?
Also I reached a dataset from a Github repository but the file format is .BIN and the required programs are a bit problematic. Here I am sharing the issue link:
https://github.com/maaghia/working-on-N2001-EMG-lab-dataset/issues/5
I found the WFDB very hard, and I am still looking for available EMG datasets.
Waiting for your answers.
- Badges
- Science topic