Systems and methods for labeling large datasets of physiological records based on unsupervised machine learning
Inventors
ARCOT DESAI, Sharanya • Tcheng, Thomas K. • Shanahan, Benjamin E.
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11842255-B2
Publication Date
2023-12-12
Expiration Date
Abstract
A deep learning model and dimensionality reduction are applied to each of a plurality of records of physiological information to derive a plurality of feature vectors. A similarities algorithm is applied to the plurality of feature vectors to form a plurality of clusters, each including a set of feature vectors. An output comprising information that enables a display of one or more of the plurality of clusters is provided, and a mechanism for selecting at least one feature vector within a selected cluster of the plurality of clusters is enabled. Upon selection of a feature vector, an output comprising information that enables a display of the record of physiological information corresponding to the selected feature vector is provided, and a mechanism for assigning a label to the displayed record is enabled. The assigned label is then automatically assigned to the records corresponding to the remaining feature vectors in the selected cluster.
Core Innovation
The invention provides a records classification processor for labeling records of physiological information. It derives a plurality of multi-dimensional feature vectors from a plurality of records of physiological information, reduces the plurality of multi-dimensional feature vectors to a corresponding plurality of reduced-dimension feature vectors, and forms a plurality of clusters, where each cluster has associated therewith a set of records of physiological information.
For at least one of the plurality of clusters, the labeling module enables an assignment of a label to each record of physiological information in the cluster set, and validates the label assigned to one or more records. Through the display and user interface, a user selects at least one record associated with a displayed cluster and assigns a label to the selected record.
The processor automatically assigns the assigned label to the other records associated with the cluster that the selected record is associated with, to create automatically-labeled records. The described workflow is directed to processing large physiological datasets and includes record selection and validation using user-assisted review.
Claims Coverage
The document includes two independent claims, one for a records classification processor and one for a method of labeling records. Across the independent claims, there are five core inventive functions: feature extraction into multi-dimensional feature vectors, dimensionality reduction to reduced-dimension feature vectors, clustering of reduced-dimension feature vectors into clusters associated with record sets, user-assisted label assignment and validation for selected records within clusters, and automatic label propagation to other records in the same cluster.
Records classification processor with feature extraction and dimensionality reduction
A feature extraction module configured to derive a plurality of multi-dimensional feature vectors from a plurality of records of physiological information; a feature reduction module configured to reduce the plurality of multi-dimensional feature vectors to a corresponding plurality of reduced-dimension feature vectors.
Clustering based on reduced-dimension feature vectors
A similarities module configured to form a plurality of clusters based on the plurality of reduced-dimension feature vectors, wherein each cluster has associated therewith a set of records of physiological information.
User-assisted label assignment and validation within clusters
A labeling module configured to enable, for at least one of the plurality of clusters, an assignment of a label to each record of physiological information included in the set of records of physiological information associated with the cluster, and a validation of the label assigned to one or more records of physiological information included in the set.
Automatic label propagation to other records in the selected cluster
The labeling module is configured to automatically assign the assigned label to the other records of physiological information associated with the cluster that the at least one selected record of physiological information is associated with, to create automatically-labeled records of physiological information.
Method of labeling records using features, reduction, clustering, and automatic propagation
Deriving a plurality of multi-dimensional feature vectors from a plurality of records of physiological information; reducing the plurality of multi-dimensional feature vectors to a corresponding plurality of reduced-dimension feature vectors; forming a plurality of clusters based on the plurality of reduced-dimension feature vectors, wherein each cluster has associated therewith a set of records of physiological information; enabling, for at least one of the plurality of clusters, an assignment of a label to each record included in the set and a validation of the label assigned to one or more records; providing an output to an interface comprising information that enables a display of the plurality of clusters; enabling a mechanism through the display and a user interface for selecting at least one record associated with one displayed cluster; providing an output to the interface comprising information that enables a display of the at least one selected record; enabling a mechanism through the display and user interface for assigning a label to the at least one displayed record; and automatically assigning the assigned label to the other records associated with the cluster that the at least one selected record is associated with, to create automatically-labeled records of physiological information.
The independent claims focus on clustering physiological records using multi-dimensional feature vectors that are reduced to reduced-dimension feature vectors, then enabling user-assisted label assignment and validation for records within at least one cluster, followed by automatically propagating the assigned label to the other records associated with that same cluster.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?