Determination of structural characteristics of an object
Inventors
Earthman, James C. • Mapar, Aboozar • Swinson, Michael David • Quan, Jr., Dennis A.
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11488062-B1
Publication Date
2022-11-01
Expiration Date
Abstract
The present invention relates generally to a system and method for measuring the structural characteristics of an object. The object is subjected to an energy application processes and provides an objective, quantitative measurement of structural characteristics of an object. The system may include a device, for example, a percussion instrument, capable of being reproducibly placed against the object undergoing such measurement for reproducible positioning. The invention provides for a system and methods for analyzing measured characteristics utilizing machine learning to create a system for predicting pathologies from measurements.
Core Innovation
The invention provides a machine learning system for evaluating structural characteristics of physical objects by using a device having an energy application tool that applies energy to an object to generate device measurements. An interface stores the device measurements, and a separate annotation user interface collects user annotations on the device measurements regarding the structural characteristics. The system combines stored measurements with user-provided structural-characteristic annotations as inputs for learning.
A training program executes a training cycle to train a machine learning algorithm on a ground truth dataset comprised of stored device measurements and annotations to create a transformation function. The transformation function is then used through a production interface to perform predictions on device measurements. The predictions evaluate structural characteristics using learned relationships between measurement data and annotated structural characteristics.
For anatomical objects, the invention further includes transforming the device measurement into a feature vector comprising an energy return versus time graph, and recording user annotations on the feature vector regarding the structural characteristics of the measured anatomical objects. Training adapts a ground truth dataset comprised of feature vector and user annotations to create the transformation function, and the production interface performs predictions on the feature vector utilizing the transformation function.
The ground truth dataset can be augmented with simulated entries generated using a physical simulation model, and the physical simulation model can be improved to better match device measurements collected. Device measurements and annotations can also be linked to information-system records identifying origin, location, or owner of the physical objects or anatomical objects measured.
Claims Coverage
This disclosure includes three independent claims that cover an end-to-end pipeline from energy-application measurement, through user annotation and ground-truth training, to a production interface that performs predictions. Across the independent claims, the core coverage consists of an energy-application measurement interface, an annotation interface, training on a ground truth dataset to create a transformation function, and prediction using the transformation function.
Machine learning system with measurement, annotation, training, and prediction interfaces
A machine learning system including a device with an energy application tool and an interface for storing device measurements, a first program logic module exposing an annotation user interface for collecting user annotations on the device measurements, a second program logic module executing a training cycle training a machine learning algorithm on a ground truth dataset to create a transformation function, and a third program logic module exposing a production interface for performing predictions on the device measurements using the transformation function.
Computer-implemented method for evaluating structural characteristics via transformation function
A computer-implemented method capturing device measurements generated using an energy application tool, annotating the device measurements with annotations regarding structural characteristics, training a machine learning algorithm with a ground truth dataset comprised of the device measurements and annotations to produce a transformation function, and applying the transformation function on captured device measurements to predict the structural characteristics.
Computerized system for anatomical objects using energy return versus time feature vector
A computerized system for evaluating structural characteristics of anatomical objects including a device that generates a measurement and transforms the device measurement into a feature vector comprising an energy return versus time graph, an annotation interface for recording a user annotation on the feature vector regarding structural characteristics, a training program executing a training cycle training a machine learning algorithm on a ground truth dataset comprised of feature vector and user annotations to create a transformation function, and a production interface for performing predictions on the feature vector utilizing the transformation function.
Overall, claim coverage centers on learning a transformation function from ground truth that pairs energy-application-derived device measurements or energy return versus time feature vectors with user annotations, and then using that transformation function in a production interface or method to predict structural characteristics for physical objects or anatomical objects.
Stated Advantages
Documented Applications
No documented applications found
Interested in licensing this patent?