Surgery – Diagnostic testing
Reexamination Certificate
2002-06-26
2004-10-19
Jones, Mary Beth (Department: 3736)
Surgery
Diagnostic testing
C600S529000, C600S544000, C128S925000, C706S924000
Reexamination Certificate
active
06805668
ABSTRACT:
FIELD OF THE INVENTION
In general, the present invention relates to monitoring patient polysomnograph data, and more specifically, to a system and method for scoring patient polysomnograph data and for providing sleep stage score feedback.
BACKGROUND OF THE INVENTION
Sleep disorders are medical problems that can cause debilitation and can be life threatening to an individual. To diagnose sleep disorders, standard practice in clinical research sleep laboratories includes the collection of physiological signals from sleeping patients. The physiological signals can include respiratory activity, blood oxygenation, brain activity, electrocardiogram (EKG), body position, leg and chin movement, etc. The data gathered from these observations can be recorded on paper or electronically for analysis. For example, a typical analysis includes examining six to eight hours of patient data in discrete time periods, such as twenty or thirty seconds, to identify key physiological factors in classifying the sleep stage of an individual.
Conventionally, a standard set of rules defined by Rechtschaffen and Kales is used by trained technicians to assign one of several unique scores, or sleep stages, to the period of time being analyzed. The sleep stages include awake, non-REM stages
1
,
2
,
3
, and
4
, and REM. Although the Rechtschaffen and Kales rules are well-defined and standardized in the relative sleep analysis industry, a typical analysis of a sleeping patient gives rise to often conflicting criteria that either satisfies more than one sleep stage, or that conflicts between two sleep stages. Accordingly, a trained technician often requires one to two hours of analysis to make a sleep stage determination. Moreover, the trained technician subjectively determines in which of the sleep stages the patient data is best classified.
In an attempt to automate sleep score analysis, some conventional computer systems have attempted to apply the Rechtschaffen and Kales rules in terms of a system implementing fixed logic, often termed “if, then” programming, for analysis. For example, the computer system would analyze the data by determining whether certain key physiological elements were present, the result of which would direct the computer system to a predetermined sleep stage score. Conventional rules-based systems can become deficient in analyzing the sleep stage of a patient, if the patient data includes conflicting factors under the Rechtschaffen and Kales scale. For example, if the system logic was created such that the presence of one factor results in the determination of a first sleep stage score, while the presence of a second factor results in the determination of a second sleep stage score, the presence of both factors in the patient data would cause the system to eliminate one of the sleep stages depending on the order in which the patient data was processed, or to otherwise not properly process the data.
A rules-based sleep processing system may be further deficient because the fixed logic is typically nonmodifiable. Because a trained technician often makes subjective decisions in processing patient data, the fixed logic also includes at least some subjective decisions inputted by the programmer into the computer system. If a user does not agree with one or more of the subjective decisions, a correction generally requires reprogramming, thereby increasing the cost of the processing system.
Another approach to utilizing computer systems to classify sleep stage scores involves generating a sleep score by utilizing a trained neural network. One skilled in the relevant art will appreciate that as applied to a system for determining a sleep stage score, a neural network system includes a plurality of inputs for accepting the patient data. Each of the inputs is processed by the neural network according to an assigned weight for each particular input. To train the neural network, the weights are adjusted after processing sample data inputs with a desired outcome. By utilizing training processes, such as backward propagation, the neural network can be trained so that the neural network will eventually generate sleep stage score output that mimics the sample data.
The use of neural networks to process patient sleep data generally facilitates the processing patient data that can include conflicting factors to achieve a patient sleep stage score. However, because a neural network relies on training from a sample set, a single neural network does not take into account differences in subjective decisions that may be arrived by two or more trained technicians. Accordingly, trained neural networks are often described as taking on the personality of the trained technician that generated the sample data. However, unlike a trained technician that is capable of providing information as, to why certain factors were considered or why a specific sleep stage score was selected, a system utilizing a single neural network can become deficient in that there is very little information regarding why the neural network generated a given result. For example, the conventional trained neural network does not output what factors were relied upon in calculating the sleep stage score. Additionally, the conventional trained neural network processing system does not indicate confidence values in how likely the calculated sleep stage score is accurate. Thus, a person, such as a trained technician, attempting to review the neural network's determination has no basis for supporting the outcome without a complete independent review.
Neural networks become further deficient in the event the patient data includes too many factors to consider. One skilled in the relevant art will appreciate that patient polysomnograph data may yield a great deal of data (e.g., more than 6,000 data points) that consumes a majority of computer processing resources. Accordingly, a neural network can abstract data by selecting more important factors within the patient data. Conventional neural networks either do not provide data abstraction, or the abstraction is not adequate, thus either consumes processing resources and/or reduces the efficiency of the neural network.
Thus, there is a need for a system and method for processing patient polysomnograph data that can automate a sleep stage score determination and that provides feedback data for review.
SUMMARY OF THE INVENTION
A system and method for processing patient polysomnograph data are provided. An abstractor obtains raw patient polysomnograph data and generates a subset of the data to include selected factors, including data clusters. The subset of the patient polysomnograph data is transferred to two or more neural networks that process the data and generate a sleep classification data. An integrator obtains the sleep classification data from the two or more neural networks by integrating the sleep classification data from each neural network. A cumulative sleep stage score is generated including confidence values and accuracy estimations for review.
In accordance with an aspect of the present invention, a system for processing patient polysomnograph data is provided. The system includes two or more neural networks operable to obtain the patient polysomnograph data and to generate a sleep classification data. The system also includes an interpreter operable to obtain the sleep classification data generated by the two or mote neural networks and to generate a cumulative sleep stage score.
In accordance with another aspect of the present invention, a method for processing patient polysomnograph data is provided. In accordance with the method, a processing system obtains the patient polysomnograph data and obtains sleep classification data corresponding to the patient polysomnograph data from a first neural network. The processing system obtains sleep classification data corresponding to the patient polysomnograph data from a second neural network and integrates the sleep classification data from the first and second neural networks to generate a cumulative sleep stage score.
In accordan
Astorino Michael
Cadwell Industries, Inc.
Christensen O'Connor Johnson & Kindness PLLC
Jones Mary Beth
LandOfFree
System and method for processing patient polysomnograph data... 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 processing patient polysomnograph data..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and System and method for processing patient polysomnograph data... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3329337