Machine learning based generation of ontology for structural and functional mapping
Inventors
Assignees
US Department of Veterans Affairs
Publication Number
US-11526808-B2
Publication Date
2022-12-13
Expiration Date
2040-05-29
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
A method may include applying, to a corpus of data, a first machine learning technique to identify candidate domains of an ontology mapping brain structure to mental function. The corpus of data may include textual data describing a plurality of mental functions and spatial data corresponding to a plurality of brain structures. A second machine technique may be applied to optimize a quantity of domains included in the ontology and/or a quantity of mental function terms included in each domain. The ontology may be applied to phenotype an electronic medical record and predict a clinical outcome for a patient associated with the electronic medical record. Related systems and articles of manufacture, including computer program products, are also provided.
Core Innovation
The invention provides a method for generating an ontology for structural and functional mapping by applying machine learning techniques to a corpus of data that includes textual descriptions of mental functions and spatial data corresponding to brain structures. The method identifies candidate domains of an ontology mapping brain circuits to mental function, where each domain corresponds to a neural circuit including one or more brain structures and associated mental function terms. A second machine learning technique optimizes the quantity of domains and mental function terms in each domain. The ontology is applied to process electronic medical records for phenotyping and predicting clinical outcomes.
The problem addressed relates to limitations in functional neuroimaging linkages between brain structures and mental functions due to conventional expert-determined knowledge frameworks. Traditional interpretations often amplify subjective biases and entrench theorized psychological constructs rather than deriving novel, replicable constructs grounded in brain function. Thus, existing brain structure-mental function links have limited novelty and replicability. The invention aims to generate an ontology in a data-driven, bottom-up manner that jointly defines domains by brain circuitry and mental functions, improving reproducibility and offering novel combinations not present in expert-determined frameworks.
Claims Coverage
The patent contains three independent claims covering a computer-implemented method, a system, and a computer-readable medium for generating and applying a machine learning-based ontology mapping brain structure to mental function.
Applying unsupervised machine learning to identify ontology domains
Applying a first machine learning technique, specifically an unsupervised learning approach such as k-means clustering on PMI-weighted co-occurrence values between brain structures and mental function terms in a corpus, to identify candidate domains in an ontology that map neural circuits including one or more brain structures to associated mental function terms.
Optimizing ontology structure with supervised forward and reverse inference models
Applying a second supervised machine learning technique involving a forward inference model trained to predict brain structure occurrence based on mental function terms and a reverse inference model trained to predict mental function terms from brain structures, to optimize the number of ontology domains and mental function terms in each domain by maximizing performance metrics such as average ROC-AUC.
Applying the generated ontology to electronic medical records for phenotyping and outcome prediction
Using the ontology to process electronic medical records by determining phenotypes based on domain ratings derived from the presence of associated mental function terms, and predicting clinical outcomes for patients, such as duration of hospital stay and emergency room visits, based on these phenotypes.
The claims cover the generation of a data-driven ontology mapping brain structures to mental functions via unsupervised and supervised machine learning techniques, and the application of this ontology to phenotyping electronic medical records and clinical outcome prediction, thereby covering the core inventive features.
Stated Advantages
Improves reproducibility of circuit-function links compared to conventional expert-determined frameworks.
Offers novel combinations of emotional and cognitive terms within ontology domains not represented in traditional frameworks.
Enhances modularity by defining internally homogeneous and distinctive domains grounded in brain circuitry and mental functions.
Generalizes well to individual articles and neurobiological phenomena, supporting representativeness of domains across human neuroimaging literature.
Enables phenotyping of electronic medical records that can predict relevant clinical outcomes, such as hospital stay duration and emergency room visits, thus potentially improving patient care and resource allocation.
Documented Applications
Phenotyping electronic medical records to determine patient characteristics based on mental functions linked to brain structures.
Predicting clinical outcomes for patients, including duration of hospital stay, quantity of office visits, emergency room visits, healthcare cost, prescriptions, refills, and comorbid conditions, based on phenotypes derived from the ontology.
Interested in licensing this patent?