Patient monitoring system with efficient pattern matching algorithm
View Patent ↗A patient monitoring system with an efficient pattern matching algorithm, a method, and a computer product thereof are disclosed. The system may include a physiological data input device or sensor which receives a plurality of physiological measurements within a time window thereby generating at least one time window data set, a memory which stores a program, and a processor. The program when executed by the processor, causes the processor to compress the at least one time window data set to a reduced-rank basis, and perform a pattern match between a reference pattern and the compressed at least one time window data set using a distance metric.
1. A method for real-time identification of a diabetes-related event in a patient using a monitoring system comprising a physiological data input device, a user input device and a processor, the method comprising:
receiving automatically from the user input device at least one reference pattern and associated alert signal;
receiving automatically from the physiological data input device at least one time window data set indicative of a physiological measurement related to the diabetes-related event;
associating automatically using the processor the at least one time window data set with a data tag;
transforming automatically using the processor the associated at least one time window data set into a normalized at least one time window data set, wherein the normalized at least one time window data set has a mean of zero and a standard deviation of one;
compressing automatically using the processor the normalized at least one time window data set into a compressed at least one time window data set;
storing automatically using the processor the compressed at least one time window data set; and
pattern matching automatically using the processor between the reference pattern and the stored at least one time window data set using a distance metric, wherein when the distance metric is less than ε, the processor automatically triggers the alert.
2. The method of claim 1 , further comprising:
transforming automatically using the processor the reference pattern into a normalized reference pattern, wherein the normalized reference pattern has a mean of zero and a standard deviation of one, and
compressing automatically using the processor the normalized reference pattern into a compressed reference pattern.
3. The method of claim 1 , wherein ε is selected so that the probability that the matches are measurements of the same physiological data is at least about 0.95.
4. The method of claim 1 , wherein ε is selected so that the probability that the matches are measurements of the same physiological data is at least about 0.98.
5. The method of claim 1 , wherein the distance metric is a Euclidean distance.
6. The method of claim 1 , wherein the compressed at least one time window data set is compressed automatically by the processor into a reduced-rank space by performing an eigen-decomposition via decomposing an {circumflex over (X)} T {circumflex over (X)} matrix into λ eigenvalues and V eigenvectors.
7. The method of claim 6 further comprising automatically calculating using the processor the cumulative sum of the eigenvalues, determining the corresponding eigenvector for each eigenvalue, and selecting a subset of eigenvectors by balancing between data compression and preservation of relevant information.
8. The method of claim 7 further comprising automatically applying using the processor an orthogonal transform matrix to said subset of eigenvectors to provide a compressed reduced-rank vector.