Classification of microwave scattering data
Inventors
Persson, Mikael • Ali Khorshidi, Mohammad • Mckelvey, Thomas • Yousefi Mesri, Hamed • Khazaeli Najafabadi, Masood
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
Example embodiments presented herein relate to solutions for analyzing and/or classifying microwave scattering data. The analyzing and/or classifying may be utilized for estimating an internal condition in an enclosed volume.
Core Innovation
The invention determines an internal condition of an enclosed volume that exhibits different dielectric properties in the presence of microwaves by using an array of transmitting antennas to emit microwaves over a frequency range and receiving antennas to receive microwave scattering measurement data scattered from the enclosed volume. The emitted and received signals define an output signal provided to a classifier with a memory that stores measurement data. The method projects the microwave scattering measurement data onto a sub-space in which a training data set lies.
The training data set comprises microwave scattering measurement data from test subjects known to be healthy. The classifier executes a mapping of the microwave scattering measurement data with respect to the training data set within the sub-space and estimates the internal condition based on that mapping. The estimated internal condition is estimated as a function of a linear distance from the projected microwave scattering measurement data and the training data set within the sub-space.
The internal condition is defined as a presence of a liquid in the enclosed volume, where the liquid changes the dielectric properties of the enclosed volume. Different distance and feature forms are described, including projecting using Higher Order Singular Value Decomposition and using an angular deviation (theta_AVRG) as an angular feature, and using subspace distance computed with an SVD/PCA-like decomposition and optional weighting via a weighting matrix (Omega).
Claims Coverage
Three independent claims are covered: a method, a device, and a non-transitory computer readable product. Across these claims, the coverage centers on microwave scattering measurement over a frequency sweep, classifier mapping to a training-data sub-space from known-healthy subjects, and estimating an internal condition as a function of distance within that sub-space for the presence of a liquid affecting dielectric properties.
Microwave scattering sensing with transmitting and receiving antenna array over frequency sweep
Using an array of transmitting antennas emitting microwaves in steps over a frequency range towards the enclosed volume, and receiving antennas in a fixed relationship to the transmitting antennas to receive microwave scattering measurement data scattered from the enclosed volume, where the received signals define an output signal provided to a classifier.
Projecting measured data onto healthy training-data sub-space
Executing a mapping of the microwave scattering measurement data with respect to a training data set comprising microwave scattering measurement data from test subjects known to be healthy, projecting the microwave scattering measurement data onto a sub-space in which the training data set lies.
Estimating internal condition as function of distance within sub-space for presence of a liquid
Estimating the internal condition based on the mapping, where the estimated internal condition is estimated as a function of a linear distance from the projected microwave scattering measurement data and the training data set within the sub-space, and the internal condition is a presence of a liquid in the enclosed volume.
Training data stored as three-dimensional complex tensor indexed by antennas and frequency
Storing the training data set in memory as a matrix or array with a dimension corresponding to at least one of a sending antenna index, a receiving antenna index, and wave frequencies utilized during a frequency sweep, where the training data set is stored as a three dimensional complex tensor with first and second dimensions corresponding to sending antenna index and receiving antenna index, and a third dimension corresponding to wave frequencies utilized during the frequency sweep.
Dielectric property determination driven by liquid-induced dielectric changes
Determining different dielectric properties of the enclosed volume due to liquid changing the dielectric properties of the enclosed volume, using the stored training data set of the test subjects as the sub-space.
Device includes antenna array, classifier, and communication port for scattered measurement data
A device comprising an array of transmitting antennas for emitting microwaves in steps over a frequency range towards the enclosed volume, receiving antennas in a fixed relationship to the transmitting antennas, a classifier, and a communication port configured to receive microwave scattered electromagnetic waves defining output signal provided to the classifier.
Temporal estimation of internal condition over a period of time
Estimating at least one internal condition over a period of time.
HOSVD-based sub-space and angular distance feature for determining internal condition
Using a Higher Order Singular Value Decomposition constructed sub-space and measuring distance as an angular distance.
Subspace distance and weighted decomposition using a weighting matrix
Using subspace distance computed using an SVD/PCA-like decomposition with an optional weighting via a weighting matrix (Omega) for determining the internal condition.
Overall, the independent claims cover microwave-scattering measurement using a transmitting and receiving antenna array over a frequency sweep, mapping and classifying the measurements relative to a healthy training-data sub-space through projection, and estimating an internal condition as a distance function within that sub-space, with training data represented as a three-dimensional complex tensor indexed by antenna indices and frequency; dependent claim refinements further specify temporal estimation and distance and feature variants including HOSVD-based angular features and subspace distance with optional weighting.
Stated Advantages
Experimental results are reported including 100% sensitivity and specificity for bleeding detection.
The document reports separable groups and threshold behavior for the classification and estimation.
Temporal monitoring is described for internal condition estimation over time.
Documented Applications
Bleeding detection using microwave scattering data with HOSVD angular features, with reported experimental results including 100% sensitivity and specificity.
Clot discrimination and temporal monitoring of internal conditions over a period of time.
Interested in licensing this patent?