Acquisition of biomedical signals databases.

Biomedical signal databases generally are built for reference purposes. In order to make use of such reference databases, a physiological or medical question first has to be specified. Like in any medical study, hypotheses have to be formulated before setting up a biomedical signal database. The questions posed and the hypotheses formulated naturally lead to the design of a study protocol. The study protocol specifies the recording equipment and details the type and number of signals and the sampling rates of the signals. The questions also allow the estimation of the number of subjects from whom the data should be collected (“power analysis”). The number of subjects and the number of total investigations then determine whether a single center can collect the data or whether a multicenter study is needed to collect the data within a reasonable period of time. One other reason supports the selection of multicenter studies. Different sites do use different equipment and follow slightly different protocols in terms of diagnosis and treatment. A study with many different partners can reflect these differences and may result being more general than it ever could be derived in a single-center study. With a thoughtful design the resulting database can answer the questions posed in the beginning. Due to the large amount of collected information, being the nature of biomedical signal databases, the database can also answer more questions by applying new analysis methodologies or hypotheses to the gathered data. But in general, any database, even with extensive data, cannot answer all questions in the field. Very often new data have to be acquired with different protocols, different channel configurations, different sampling rates, or different signal preconditioning, etc. As biomedical signal databases are used for scientific and reference purposes, one specific aim is to obtain the best signal quality possible. Therefore, quality assurance during data acquisition is a very important task and has to follow structured guidelines. However, there is a lack of such general guidelines and often they need to be database specific at least to some extent. Often, signal databases are created in multicenter studies. Multicenter studies require additional specifications in terms of recording equipment and protocol compatibility. Special care must be taken for continuous quality assurance during the recording period at the different sites. Site visits at all recording laboratories are a very useful method to harmonize recording conditions. Such site visits should follow a checklist of items being tracked. Although the checklist can be planned ahead of the recordings, its should be revised after the first round of recordings from each participating laboratory. This test series often reveals unforeseeable deviations from the intended protocol. Usually, the first objective of recording biomedical data is not to make a database but to monitor or diagnose a patient, or to do research. The recorded data strongly depend on this first objective. For instance, monitoring the vital signs of a patient in acute life-threatening states or being under surgical procedures or anesthesia conditions requires online analysis and immediate visualization. This is needed to allow immediate intervention by personnel trained for such situations. When data is recorded for research or diagnosis, then immediate visualization of analyzed data is usually not required, but continuous storage of data is. Annotations and expert scorings of the signals recorded are as important as the data itself. Only through the evaluation of