Working with the EDF+ standard for polysomnographic analyses, I’ve noticed some limitations of the format such as the low standardization of annotation labeling scheme and the fact that adding annotations sometime can requires rewriting the whole data file, which is quite inefficient. I’m very interested in knowing what the limitations that researchers using this standard encountered are and how they have get around these limitations. For example, for the second limitation I gave as example, I spited the recordings and the annotations in two separate EDF+ files such that modifications to the annotations never require changing the recording file.
May I suggest you to contact with Dr. Alvarez-Estevez ([email protected])? He probably will know the best answer to your question.
You may also contact Bob Kemp ([email protected]) who has developed the EDF format.
Can I suggest collaboration with the on-going AANEM-standards to be incorporated in EDF+-formatting standards for PSG? They are reaching very high levels of inter-rater reliability in PSG scoring, with a concept that will only improve it.
Thanks, Diego Alvarez-Estevez, for informing me about these questions of Christian and Antoinette.
1. Christian, have you seen the list of standard texts at http://www.edfplus.info/specs/edftexts.html? What kind of extra standardization would you suggest?
2. Christian, in my sleep lab, the PSG file contains the recorded signals and only the on-line annotations, either on-line typed by the technician or coming from a marker button. We do not allow any change to that file, so it is never rewritten. Any annotations made off-line after the recording session (possibly made by a different technician), such as sleep scores but also corrections of the on-line annotations, are put in a separate file. If a third technician would also score the PSG, she produces her own annotations file.
We typically record PSGs with 13 EXG signals sampled at 256Hz, 11 polygraphic signals (resp, SaO2, sound, light, body position etc) sampled at different frequencies of 4 - 256Hz. The file has datablocks that are each about 10kByte in size and 2 seconds in duration. The annotations have 120 samples, that is 240 bytes, reserved in each datablock. Not ever in a history of many thousands of EDF+ recordings did we need more than 240 bytes in 2 seconds. No technician types that fast. Rewriting the big PSG files (typically 500MByte) is never necessary.
The off-line file contains only annotations including the sleep scores, respiratory events, leg movement events and so on. These files are typically 50kByte till 100kByte, so rewriting those is no problem. In fact they are rewritten as a backup several times during the scoring procedure without the technician noticing that.
Your suggestion to also save the on-line annotations separate from the PSG signals is interesting. It makes life for the programmers more easy, at the rather small cost of having to organize the coupling between those two files. We did not choose this option because we wanted to disable changing the on-line obtained annotations. Having the on-line annotations in a separate file (just like the off-line annotations) would make it more easy to change those unless the on-line file is protected in another way.
3. Antoinette, can you provide info about the AANEM standards? The standardization of EEG terms has until now mainly been limited to electrode names and polarization rules. The EDF+ article explicitly notes the possibility to add other standard texts.
Thank you very much Dr. Kemp for your answer. It is very insightful to know how you use the EDF+ format in your lab. The context of my research center is a little different. We use commercial software that records the data in a proprietary format. However, we are about to release an open-access database of PSG to help stimulate the development of PSG signal analysis tools. For that purpose, I’ll be converting our data to EDF. Thus, the “online versus offline” difference is not a matter in this scenario, but I do want to separate data from annotations for reasons that were previously exposed.
For your question regarding possible kind of extra standardization, one of the limitations I’ve encountered using EDF+ is that I like to be able to associate features with events. For that purpose, I use a XML string as annotation. For example, the following string would describe a spindle event occurring on channel Cz during sleep stage 2 of the second REM-NREM cycle with a root mean square (RMS) amplitude of 12.07 uV: . I think such a formatted annotation can be of great use but it requires some standardization to allow inter-operability. This is even more important in the context of the database we will be releasing. Through a “web 2.0” page, we will invite researchers to collaborate with submitting their own annotations to these nights. For example, agreeing in a common standard for the names of these fields is very important in this situation. Also, since I feel that I cannot anticipate the needs of everybody, I’ll invite researchers using or collaborating to this database to submit suggestions of improvement/modifications to this annotation scheme. Because it is some kind of subset of the EDF+ standard (much as EDF+ is for EDF), I would be tempted to refer to this evolving format as EDF++ merely to distinguish it and to avoid confusion (don’t want to steal your name though). I hope this further step in standardization of the base format can evolve naturally from the online exchange with the researchers in the domain. Your contribution at any step of the process will of course have the uttermost value.
One more thing I’m considering at the moment, is adopting a 24-bit resolution for ouy database just as for the BDF format. I think memory is becoming a less and less an important factor while increased accuracy is valuable.
Antoinette: I’m also interested in your comment on the AANEM standards. I would also be interested by more info on this…
- Badges
- Science topic
- Similar topics
- Correction