The Montreal Archive of Sleep Studies (MASS) is an open-access and collaborative database of laboratory-based polysomnography (PSG) recordings [1-1] . Its goal is to provide a standard and easily accessible source of data for benchmarking the various systems developed to help the automation of sleep analysis. It also provides a readily available source of data for fast validation of experimental results and for exploratory analyses. Finally, it is a shared resource that can be used to foster large-scale collaborations in sleep study.
MASS is composed of cohorts themselves comprising subsets. Recordings within subsets is kept as homogeneous as possible, whereas it is more heterogeneous between subsets. To allow inter-study comparisons, researchers validating their results on MASS are encouraged to specify which portion of the database they used in their assessment (e.g., MASS-C1 for the whole cohort 1, MASS-C1/SS1-SS3 for subsets 1, 2 and 3 of cohort 1).
Currently, the following cohorts are available: Cohort 1
August 1, 2015
Due to a bug in the exportation program from Stellate Harmonie native file format (.sig/.sts) to EDF+ file format, some channels have been dropped off from the final database released in 2014. More importantly, it resulted in the step involving the recomputation of the reference electrode to be skipt silently, meaning that the original references used for recording were conserved in these files instead of the recomputed ear-linked references. This does not affect SS3, SS5 and some files of SS1 which were referenced to ear electrodes linked with a 10 kOhm resistor during recording. However, the other files remained with their original references which was either A1 or A2 (recording references are listed in open-access descriptors available here ). These issues have been corrected in the database, as available from August 1, 2015. The 2014 version of the database can be obtained upon request. Here is a list of the papers that have been published using the original version of the database (if your publication should be added here, please let us know ): [2-1 ], [2-2 ], [2-3 ], [2-4 ].
This change is expected to be of negligeable consequences for studies concerned with automated scoring. For example, we tested the performance of one of our spindle detectors on SS2 recordings for the old version (2014) and the new version (2015).
On this figure E1 and E2 are referring to the two sets of expert scorings. As shown, the Matthew’s correlation coefficient (MCC) assessing the performance of the detector takes very similar values for the two versions. Paired Student t-test showed non-significant differences (p > 0.05) for MCC values computed with the old and with the new versions.
We nevertheless recommend that any new publications use the updated version of the database. We are apologizing for any inconvenient this issue may have caused.
Questions related to MASS should be addressed