1、TechnicalInformationReportAAMI TIR No. 24:1999AAMIAssociation for theAdvancement of MedicalInstrumentationAcquisition and use ofphysiologic waveform databasesfor testing of medical devicesAAMI TIR No. 24:1999Acquisition and use ofphysiologic waveform databasesfor testing of medical devicesApproved 1
2、8 August 1999 byAssociation for the Advancement of Medical InstrumentationAbstract: This report defines the nomenclature, ingredients, and principles needed to develop, annotate,evaluate, and use physiologic waveform databases in developing and testing medical devices.Keywords: waveform, physiologic
3、, algorithms, arrhythmiaPublished byAssociation for the Advancement of Medical Instrumentation1110 N. Glebe Road, Suite 220Arlington, VA 22201-4795 2000 by the Association for the Advancement of Medical InstrumentationAll Rights ReservedPublication, reproduction, photocopying, storage, or transmissi
4、on, electronically or otherwise, of all or any part of thisdocument without the prior written permission of the Association for the Advancement of Medical Instrumentation isstrictly prohibited by law. It is illegal under federal law (17 U.S.C. 101, et seq.) to make copies of all or any part ofthis d
5、ocument (whether internally or externally) without the prior written permission of the Association for theAdvancement of Medical Instrumentation. Violators risk legal action, including civil and criminal penalties, anddamages of $100,000 per offense. For permission regarding the use of all or any pa
6、rt of this document, contactAAMI, 1110 N. Glebe Road, Suite 220, Arlington, VA 22201-4795. Phone: (703) 525-4890; Fax: (703) 525-1067.Printed in the United States of AmericaISBN 1570201323AAMI Technical Information ReportA technical information report (TIR) is a publication of the Association for th
7、e Advancement of MedicalInstrumentation (AAMI) Standards Board that addresses a particular aspect of medical technology.Although the material presented in a TIR may need further evaluation by experts, releasing the information isvaluable because the industry and the professions have an immediate nee
8、d for it.A TIR differs markedly from a standard or recommended practice, and readers should understand the differencesbetween these documents.Standards and recommended practices are subject to a formal process of committee approval, public review, andresolution of all comments. This process of conse
9、nsus is supervised by the AAMI Standards Board and, in the caseof American National Standards, the American National Standards Institute.A TIR is not subject to the same formal approval process as a standard. However, a technical committee and theAAMI Standards Board do approve a TIR for distributio
10、n.Another difference is that although both standards and TIRs are periodically reviewed, a standard must be acteduponeither reaffirmed, revised, or withdrawnand the action must be formally approved usually every 5 years butat least every 10 years. For a TIR, AAMI consults with a technical committee
11、about every 5 years after thepublication date (and periodically thereafter) for guidance on whether the document is still usefulthat is, to checkthat the information is relevant or of historical value. If the information is not useful, the TIR is removed fromcirculation.A TIR may be developed becaus
12、e it is more responsive to underlying safety or performance issues than a standardor recommended practice or because achieving consensus is extremely difficult or unlikely. Unlike a standard, a TIRpermits the inclusion of differing viewpoints on technical issues.CAUTION NOTICE: This AAMI technical i
13、nformation report may be revised or withdrawn at any time. Because itaddresses a rapidly evolving field or technology, readers are cautioned to ensure that they have also consideredinformation that may be more recent than this document.ContentsPageCommittee representation viiAcknowledgments viiForew
14、ord. viiiIntroduction .ix1 Scope12 Normative reference .13 Definitions.14 Database requirements 24.1 Intended use of database 24.2 Clinical requirements .24.2.1 Population24.2.2 Study design24.3 Engineering requirements24.4 Annotation requirements34.5 Archive requirements.35 Waveform acquisition and
15、 synthesis 35.1 Overview 35.2 Signal issues and requirements.35.2.1 Distortion .35.2.2 Distortion classifications 45.2.3 Skew65.2.4 Duration .65.2.5 Frequency translation 65.3 System design issues and requirements .65.3.1 Sampling theory considerations 75.3.2 Architectural issues .215.3.3 Channel ac
16、quisition guidelines225.4 Storage 235.4.1 Raw signalsanalog databases .245.4.2 Waveformsdigital databases245.5 Archive.245.5.1 Environmental considerations .245.5.2 Format .255.5.3 Data integrity255.5.4 Applicable standards .255.6 Database annotation255.6.1 Annotation process255.6.2 Annotation rules
17、.265.6.3 Data presentation specifications265.7 Maintenance and distribution.276 Application of waveform databases to testing 276.1 Evaluation of performance.276.2 Test objectives.276.3 Algorithm versus device testing system.286.4 Sufficiency and validity.286.4.1 Sufficiency .286.4.2 Validitycorrect
18、conclusions from results.296.4.3 Precautions, limitations, interpolations, and interpretations (limited scope of results) 29AnnexesA CSE ECG Reference Library (Measurement Database).31B CSE ECG Reference Library (Diagnostic Database)33C MIT-BIH Arrhythmia Database36D Noise Stress Database .38E Europ
19、ean ST-T Database.39F American Heart Association Database forEvaluation of Automated Ventricular Arrhythmia Detectors41G Massachusetts General Hospital/Marquette Foundation Waveform Database 42H Creighton University Ventricular Tachyarrhythmia Database .43Table1 Minimum fs/BW ratios vs. A.A. filter
20、order.23Figures1 Sampled signal spectra and aliasing 82 Ideal vs. practical anti-alias filter characteristics.93 Anti-alias filtering for signal oversampling 94 Anti-alias filtering for sigma-delta converters105 Monochromatic, 20-Hz sine wave sampled at 500 samples per second106 Amplitude spectrum f
21、or figure 5117 Effect of irregular sampling of a monochromatic waveform128 Amplitude spectrum for figure 7129 Improper sample rate reduction1410 Improper sample rate reduction1511 Improper sample rate expansion of a 2-Hz sine wave sampled at 60 s/s expanded to 300 s/sand as if it were latched 1612 I
22、mproper sample rate expansion 1613 Proper sampled rate reduction .1814 Proper sampled rate reduction .19 2000 Association for the Advancement of Medical Instrumentation ! AAMI TIR No. 24:1999 viiCommittee representationAssociation for the Advancement of Medical InstrumentationWaveform Testing Commit
23、teeThe AAMI Waveform Testing Committee developed this technical information report (TIR). Committee approval ofthe TIR does not necessarily imply that all committee members voted for its approval.At the time this document was balloted, the AAMI Waveform Testing Committee had the following members:Co
24、chairs: Carl A. PantiskasSandy WeiningerMembers: James J. Bailey, MD, National Institutes of HealthDon Brodnick, Marquette MedicalRichard Diefes, ECRI-MSLBRobert Donehoo, Critikon/Johnson at least two times thehighest frequency component of the signal.2 2000 Association for the Advancement of Medica
25、l Instrumentation ! AAMI TIR No. 24:19993.12 oversampling: An analog I/O (input oversampling) technique whereby samples are either read in orreproduced at a rate equal to some constant (the interpolation factor) times the fundamental sample rate.NOTEThis technique is used to suppress input and outpu
26、t aliasing and has no bearing on ratios between signal frequencycomponents and the recordings sample rate. It can be more clearly stated as either input oversampling or output oversampling.3.13 quantization error: One-half the analog value range associated with the least significant bit in a system.
27、NOTEFor example, a 12-bit converter has 4,096 discrete states possible. If it is applied to a signal with a 5-volt range, there isa step size equal to 10 times the total voltage range divided by 4,096, for nearly 2.44 mV per step. Thus, any analog voltage within 1.22 mV of the center of each value r
28、ange will be associated with exactly the same digital representation. This uncertainty isassociated with finite precision quantization. Its range is twice the quantization error and is sometimes referred to as the uncertaintybit size.3.14 power spectrum: A representation of the energy content per un
29、it frequency. The accuracy of the estimationof a populations power spectrum depends on an ensemble average.3.15 SNR: Signal-to-noise ratio.3.16 sample rate: The rate at which the instantaneous value of an analog signal is quantized to one of the Ndiscrete voltage levels an ADC can resolve.3.17 VF: V
30、entricular fibrillation.4 Database requirementsThe following requirements have implications for database development and the use of databases use in assessingmedical device performance, and should be considered when planning to build a database.4.1 Intended use of databaseThe intended use of a wavef
31、orm database determines its requirements and system specifications. For example, adatabase that is intended for training purposes needs only enough fidelity so that users can interpret the waveforms.A higher-fidelity recording system may be needed for a database that will be used for algorithm testi
32、ng withoutreconstruction to the analog domain. If the database is used for system verification testing, then the demands on thefidelity of the recording and simulation system may be much higher. In a similar manner, the requirements fordatabase annotation may be quite different for each intended use
33、. A mismatch between the intended use and theresultant requirements may produce a system whose testing results are invalid or whose costs far exceed thebenefits derived.4.2 Clinical requirementsIt is impractical to collect data until every expected event has been captured. Many times, it is sufficie
34、nt to synthesizewaveforms to augment the recorded physiologic data. This approach should be discussed before databasedevelopment to facilitate appropriate decisions at each stage.4.2.1 PopulationInclusion and exclusion criteria must be established for each clinical variable and must be traceable to
35、thedatabases intended use. For example, using a database of adult signals to assess the performance of infantmonitors is incorrect.4.2.2 Study designThe study design depends on how the patients are stratified with respect to risk or disease categories. An importantdesign problem is how to collect a
36、sufficient number of cases in each category so that classification algorithms canbe expected to achieve reasonable statistical power and accuracy. Further, if the design cannot identify signals thatare pathognomonic for each category, then signal characteristics must, at least, be distinctive enough
37、 to separatethe categories of patients, again, with reasonable statistical power. The design should anticipate whether parametersthat are extracted from the signals will be used in a stepwise, discrete fashion or as continuous variables.4.3 Engineering requirementsThe characteristics of each signal
38、(e.g., dynamic range, resolution, mode floor, measurement precision, andprecision of temperature measurement) to be collected must be specified. The duration of data collection must bespecified. Special needs in the clinical setting that may affect the architecture of the equipment must be identifie
39、d. 2000 Association for the Advancement of Medical Instrumentation ! AAMI TIR No. 24:1999 34.4 Annotation requirementsThere must be a set of rules for analyzing the recorded physiological signals, a method of administering the rules,and a system for incorporating the results into a waveform database
40、. These rules guide the referee annotators tomark and characterize events of interest in the recorded data; these a priori rules must be specifically tailored to thedatabase. The data in the database should not be used to determine the rules. The annotation methodology andimplementation may place co
41、nstraints on the engineering of the signal acquisition system and must be establishedin the planning stage before any development.4.5 Archive requirementsThe methodology for archiving the data should be considered in the planning stage before any data collection. Thearchive should also contain docum
42、entation of the software and hardware data acquisition systems, data formats, andmethods for recovering data. Additionally, the archive database storage, maintenance, and ownership should also beaddressed.5 Waveform acquisition and synthesisThis section describes the procedures for recording wavefor
43、ms within the database and developing a system to playback the waveforms.5.1 OverviewAny database use strongly depends on the quality of the equipment that is used to record (acquire), store, andreproduce the physiological signals of interest. At each step of the process, care must be taken to ensur
44、e that thesignals integrity is maintained, particularly for databases that ultimately will be used to simulate a patient. In short,the signal that is recorded and stored in a database must be reproducible with sufficient accuracy to ensure that thedevice or algorithm to be tested is presented with a
45、n adequate representation of the original physiologic signal.The validity of a multiparameter monitor test method depends on the capability to record (acquire) and reproducehigh-quality physiologic signals without introducing significant error. As mentioned previously, acquisition fidelityshould mat
46、ch the intended use of the database. A high-fidelity acquisition system would be required for reproducingphysiologic signals from CD-ROM media to drive the transducer(s) of a monitor under test. However, a lower-fidelityrecording system may be designed if the waveforms will be used only for display
47、during training exercises. In eithercase, care must be taken at each step of the design process to ensure appropriate signal integrity.The overall architecture of the databases data acquisition system profoundly affects the accuracy of the collecteddata and the reproduced waveforms during playback.
48、Therefore, two initial tasks required to develop and to use aphysiologic database are to determine the general signal requirements and to establish system specifications tomeet them.5.2 Signal issues and requirementsThis section discusses the significance of general signal issues, the requirements o
49、f system design, and their effecton specifications of the systems signal acquisition, storage, and reproduction subsystems. Sufficient backgroundtheory is given where needed to justify the selected implementation.Four signal acquisition and reproduction issues have been identified as being important for testing physiologicmonitors and algorithms: distortion, relative timing (skew), maximum uninterrupted duration, and frequencytranslation.5.2.1 DistortionThe most important design goal for a waveform recording and playback system is preserving the signalswaveshape. Kn
