Adaptive prediction of changes of physiological/pathological states using processing of biomedical signals
First Claim
1. Method for predicting changes of physiological/pathological states in a patient, based on sampling, processing and analyzing a plurality of aggregated noisy biomedical signals, comprising:
- a) generating a reference database of raw data streams or features, derived from said raw data streams, representing physiological/pathological states, by aggregating one or more raw data streams, each of which consisting of biomedical signals of a plurality of patients, at least several of which having one or more of said physiological/pathological states, wherein said features are obtained by performing, for each type of biomedical signal, adaptive segmentation of its corresponding raw data streams, and assigning, to each segment, individual attributes being represented by attribute values, thereby obtaining data related to each physiological/pathological states;
b) generating additional data streams using said attributes;
c) Repeating steps a), b) if needed;
d) determining an attribute domain, in which each segment being represented by a point that corresponds to the attribute values of said segment;
e) for each physiological/pathological state, generating a set of clusters in said attribute domain, each of which consisting of a combination of points determined by their relative location to other points, by assigning a set of property values to each point, each property value corresponding to the degree of association of said point with one of the clusters;
f) associating each point, in time, to a corresponding state;
g) determining the probabilities of transitions between states by obtaining the frequency and the order of appearance of each point, in time;
h) repeating steps d) to f) above while in each time, varying the combination of points included in each cluster according to their most updated property value and by including points derived from said probability until said updated property values remain essentially unchanged, thereby updating each cluster and said probabilities of transitions;
i) generating prior knowledge data, consisting of a plurality of feasible paths between states according to said probabilities of transitions, by associating each feasible path with a corresponding dynamics of transitions between physiological/pathological states;
j) associating at least one updated cluster with a normal/abnormal physiological state of said patient by using former knowledge, regarding normal/abnormal physiological/pathological states;
k) For each patient,l) aggregating one or more individual data streams or features, derived from said individual data streams, each of which consisting of biomedical signals of said patient, wherein said features are obtained by performing, for each type of biomedical signal, adaptive segmentation of its corresponding raw data streams, and assigning, to each segment, individual attributes being represented by attribute values;
m) Forming additional data streams out of said attributes;
n) Repetition of stages k), l) if needed;
o) assigning each individual attribute to a corresponding state, or to a new state, according to the probability to belong to each existing cluster or to a new cluster associated with said new or existing state and said probabilities of transitions;
p) adaptively updating each existing or new cluster and said probabilities of transitions according to said individual data streams;
q) obtaining a path, being an individual dynamics, between physiological/pathological states according to their order of appearance; and
r) obtaining a prediction of being in, or transitions to, physiological/pathological states in said patient, by comparing said individual dynamics with known dynamics, obtained from prior knowledge.
2 Assignments
0 Petitions
Accused Products
Abstract
A method and system predicts changes of physiological/pathological states in a patient, based on sampling, processing and analyzing a plurality of aggregated noisy biomedical signals. A reference database of raw data streams or features is generated by aggregating one or more raw data streams. The features are derived from the raw data streams and represent physiological/pathological states. Each feature consists of biomedical signals of a plurality of patients, wherein several patients have one or more of the physiological/pathological states. A path, which is an individual dynamics, between physiological/pathological states is obtained according to their order of appearance. Then, a prediction of being in physiological/pathological states, or transitions to physiological/pathological states in the patient, is obtained by comparing the individual dynamics with known dynamics, obtained from prior knowledge.
368 Citations
46 Claims
-
1. Method for predicting changes of physiological/pathological states in a patient, based on sampling, processing and analyzing a plurality of aggregated noisy biomedical signals, comprising:
-
a) generating a reference database of raw data streams or features, derived from said raw data streams, representing physiological/pathological states, by aggregating one or more raw data streams, each of which consisting of biomedical signals of a plurality of patients, at least several of which having one or more of said physiological/pathological states, wherein said features are obtained by performing, for each type of biomedical signal, adaptive segmentation of its corresponding raw data streams, and assigning, to each segment, individual attributes being represented by attribute values, thereby obtaining data related to each physiological/pathological states; b) generating additional data streams using said attributes; c) Repeating steps a), b) if needed; d) determining an attribute domain, in which each segment being represented by a point that corresponds to the attribute values of said segment; e) for each physiological/pathological state, generating a set of clusters in said attribute domain, each of which consisting of a combination of points determined by their relative location to other points, by assigning a set of property values to each point, each property value corresponding to the degree of association of said point with one of the clusters; f) associating each point, in time, to a corresponding state; g) determining the probabilities of transitions between states by obtaining the frequency and the order of appearance of each point, in time; h) repeating steps d) to f) above while in each time, varying the combination of points included in each cluster according to their most updated property value and by including points derived from said probability until said updated property values remain essentially unchanged, thereby updating each cluster and said probabilities of transitions; i) generating prior knowledge data, consisting of a plurality of feasible paths between states according to said probabilities of transitions, by associating each feasible path with a corresponding dynamics of transitions between physiological/pathological states; j) associating at least one updated cluster with a normal/abnormal physiological state of said patient by using former knowledge, regarding normal/abnormal physiological/pathological states; k) For each patient, l) aggregating one or more individual data streams or features, derived from said individual data streams, each of which consisting of biomedical signals of said patient, wherein said features are obtained by performing, for each type of biomedical signal, adaptive segmentation of its corresponding raw data streams, and assigning, to each segment, individual attributes being represented by attribute values; m) Forming additional data streams out of said attributes; n) Repetition of stages k), l) if needed; o) assigning each individual attribute to a corresponding state, or to a new state, according to the probability to belong to each existing cluster or to a new cluster associated with said new or existing state and said probabilities of transitions; p) adaptively updating each existing or new cluster and said probabilities of transitions according to said individual data streams; q) obtaining a path, being an individual dynamics, between physiological/pathological states according to their order of appearance; and r) obtaining a prediction of being in, or transitions to, physiological/pathological states in said patient, by comparing said individual dynamics with known dynamics, obtained from prior knowledge. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23)
-
-
24. System for predicting changes of physiological/pathological states in a patient, based on sampling, processing and analyzing a plurality of aggregated noisy biomedical signals, comprising:
-
a) Data acquisition means for collecting biomedical signals of one or more patients; b) A database of data streams or features, derived from said data streams, representing physiological/pathological states, said database aggregates one or more raw data streams, each of which consisting of said biomedical signals of a plurality of patients, at least several of which having one or more of said physiological/pathological states, said database being capable of storing data streams or features which are used as reference data streams or features for characterizing further individual patients and for storing data streams or features of individual patients; c) First processing means for obtaining said features by performing, for each type of biomedical signal, adaptive segmentation of its corresponding raw data streams, and assigning, to each segment, individual attributes being represented by attribute values, and for obtaining data related to each physiological/pathological states; and d) Second processing means for determining an attribute domain, in which each segment being represented by a point that corresponds to the attribute values of said segment;
for generating a set of clusters in said attribute domain for each physiological/pathological state wherein each of which consisting of a combination of points determined by their relative location to other points, for assigning a set of property values to each point, wherein each property value corresponding to the degree of association of said point with one of the clusters;
for associating each point, in time, to a corresponding state;
for determining the probabilities of transitions between states by obtaining the frequency and the order of appearance of each point, in time;
for varying the combination of points included in each cluster according to their most updated property value and for including points derived from said probability until said updated property values remain essentially unchanged, so as to update each cluster and said probabilities of transitions;
for generating prior knowledge data, consisting of a plurality of feasible paths between states according to said probabilities of transitions, by, and for associating each feasible path with a corresponding dynamics of transitions between physiological/pathological states;
for associating at least one updated cluster with a normal/abnormal physiological state of said patient by using former knowledge, regarding normal/abnormal physiological/pathological states;
for performing, for each type of biomedical signal, adaptive segmentation of its corresponding raw data streams, and for assigning, to each segment, individual attributes being represented by attribute values, for assigning each individual attribute to a corresponding state, or to a new state, according to the probability to belong to each existing cluster or to a new cluster associated with said new or existing state and said probabilities of transitions;
for adaptively updating each existing or new cluster and said probabilities of transitions according to said individual data streams;
for obtaining a path, being an individual dynamics, between physiological/pathological states according to their order of appearance, and associating said most feasible path with an individual dynamics of transitions between physiological/pathological states; and
for obtaining a prediction of being in, or transitions to, physiological/pathological states in said patient, by comparing said individual dynamics with known dynamics, obtained from prior knowledge. - View Dependent Claims (25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46)
-
Specification