Data processing: measuring – calibrating – or testing – Measurement system – Performance or efficiency evaluation
Reexamination Certificate
2000-06-16
2003-04-29
Bui, Bryan (Department: 2863)
Data processing: measuring, calibrating, or testing
Measurement system
Performance or efficiency evaluation
C702S182000, C702S186000, C706S015000
Reexamination Certificate
active
06556951
ABSTRACT:
FIELD OF THE INVENTION
The invention relates generally to monitoring processes to insure their proper functioning. Such processes include (but are not limited to) acquisition of medical specimen data from laboratory instruments, assembly line manufacturing, and general plant or factory operation. More particularly, the invention describes a two-tiered automated intelligent agent architecture for real-time monitoring of measured values in the process environment to detect and correct unusual and untoward events including catastrophic failure. A secondary use of this invention is as a method for enhancing both human and machine learning, recognizing, and diagnosing untoward events in the process environment as signaled by time-varying behavior patterns of measured values of relevant parameters.
BACKGROUND OF THE INVENTION
Process control ensures that products of a process, whether the products are physical objects or information, are produced correctly. Central to process control is monitoring and assessing measured values in the process, which can be achieved by various computational methods whenever features of a product being produced can be quantified. Through such oversight the process can be corrected to prevent the production of defective products. The problem, however, is how to monitor the process effectively where errors in the process are not readily detectable from the individual products produced (e.g., a laboratory test result). Ideally, the monitoring would detect an error in the process immediately after its appearance, before few defective products are produced. But because of ineffective monitoring, the time between appearance of a process error and its detection is often too long and many defective products may be produced before the error is detected.
For example, the analysis of patient specimens by various clinical laboratory instruments such as chemical analyzers is susceptible to processing errors. These errors can be of several types. Inaccuracy, or analytical bias, occurs if the measured value of patient data systematically deviates from its true value. Imprecision occurs if there is a systematic increase in variability about the true value. And random error occurs when the measured value of patient data differs from the true value non-systematically. Unless these errors are promptly detected and corrected, physicians may receive inaccurate test results, which can lead to misdiagnosis and inappropriate patient management.
A common monitoring procedure for ensuring that clinical laboratory instruments are working properly is to perform all analyses in “batch” mode. (A “batch” is often referred to as a “run”.) Patient specimens that are received in the laboratory are saved until enough are collected for a run. The accuracy of the instruments is tested before and after the run by assaying the run control materials which contain known concentrations of the substances being measured. If instrument error is detected, the instruments are adjusted and the specimens are again analyzed before test results are reported.
Batch mode monitoring ensures accurate results, but it is too slow for modern clinical laboratory practice. Patient specimen analysis is frequently required immediately and cannot be postponed until enough other patient specimens are collected for a run. Furthermore, it is not cost effective to test the instruments before and after analyzing individual patient specimens.
Another monitoring approach is to test laboratory instruments periodically (such as every six to 12 hours) with the control materials and adjust the instruments as necessary. This approach allows prompt analysis of patient data, but has the drawback that some patient specimens may be inaccurately analyzed and the results reported before errors are detected.
An alternative to monitoring the quality of laboratory instruments with control material is to monitor them using the test results they generate (see Hoffman, R. G. et al., “The Average of Normals Method of Quality Control,”
Am. J. Clinical Pathology,
vol. 105, pp 44-51 (1996)). But except for the daily monitoring of mean test result data, such monitoring is not widely used because current computational methods have limited ability to detect instrument errors from test result data (see Dembrowski, G. S. et al., “Assessment of Average of Normals Quality Control Procedures and Guidelines for Implementation,”
Am. J. Clinical Pathology,
vol. 81, pp 492-99 (1984)). The problem with monitoring the daily mean is that it is a retrospective detection method with a long time frame. Potentially a full day of incorrect test results may be reported before such a method detects an instrument failure.
One objective of the invention, therefore, is to provide a practical and cost-effective method and system for detecting errors as they appear (i.e., “in real time”) in a general processing environment, before many defective products are produced. An example of such a process to which the invention can be applied is the performance of a laboratory instrument in analyzing patient specimens. Another objective of the invention is to provide such a method and system that utilizes computational methods to analyze process data for identifying the presence of various types of process errors. In the application of this invention to the medical field, problems with laboratory instruments are discovered and corrected before a significant amount of patient data is affected by monitoring test results as they are produced.
SUMMARY OF THE INVENTION
The invention comprises a computer-implemented method for detecting errors in a process such as clinical laboratory data production, assembly line manufacturing, and general assembly plant or factory operations. The method includes the following steps. Data elements having a range of values are collected from the process. The number of data elements having values within predetermined intervals of the range are then counted. The counts of the data elements are applied as inputs to nodes of a neural network, each count being applied to a node representing the predetermined interval corresponding to the count. Output indicative of whether an error has occurred in the process is then generated from the neural network based on the inputs.
In a more specific form of the invention, the step of collecting data elements comprises collecting through a moving window a predetermined number of data elements. The steps of the method are then repeated after moving the window to replace a data element in the window with another newly produced data element outside the window.
In another more specific form of the invention, the method includes the additional step of scaling the data elements to values within a predetermined range before counting the data elements. This scaling step may comprise scaling the data elements to values within the range of zero to one, for example.
In another more specific form of the invention, the neural network is trained to detect a bias error or a precision error in the data elements applied as inputs to the nodes.
Although the invention is applicable to many processes, it is of particular value for detecting errors in the laboratory analysis of patient specimens. When applied to this process, the collecting step comprises collecting test results generated by a laboratory instrument processing patient specimens. The test results are the data elements that are counted, and the counts are applied to the neural network. The neural network monitors trends in test results to detect if the instrument is producing error-free test results. If test results outside of an acceptable range are detected, the neural network output indicates that an error has occurred. The output is generated in real time and available immediately for automatic or manual correction of the instrument. Thus the method substantially eliminates the delay inherent in prior approaches for detecting and correcting laboratory instrument error.
The invention also comprises a system that implements the method.
The foregoing and other objec
Deleo James M.
Remaley Alan T.
Bui Bryan
Klarquist & Sparkman, LLP
The United States of America as represented by the Secretary of
LandOfFree
System and method for intelligent quality control of a process does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with System and method for intelligent quality control of a process, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and System and method for intelligent quality control of a process will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3065463