System and method for orientating capture of ultrasound images
Inventors
KEZURER, Itay • Eshel, Yoram • Ludomirsky, Achiau • Lipman, Yaron
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12450483-B2
Publication Date
2025-10-21
Expiration Date
Abstract
A device for a mobile ultrasound unit having an ultrasound probe includes a trained orientation neural network and a result converter. The trained orientation neural network receives a non-canonical image of a body part from the mobile ultrasound unit and generates a transformation transforming between a position and rotation associated with a canonical image and a position and rotation associated with the non-canonical image. The result converter converts the transformation into position and/or rotation instructions of the probe and displays the position and/or rotation instructions to a user to change the position and/or rotation of the probe.
Core Innovation
The invention relates to a mobile handheld ultrasound navigator that receives a non-canonical image of a body part from an ultrasound unit and generates a transformation between a position and rotation associated with a canonical image and a position and rotation associated with the non-canonical image. A result converter converts the transformation into position and/or rotation instructions of the probe and displays the position and/or rotation instructions to a user to change the position and/or rotation of the probe. The navigator further includes an orientation neural network that is trained to generate the transformation from the received non-canonical image.
The invention addresses training of the orientation neural network using canonical and non-canonical images taken around the canonical image together with transformations between positions and rotations in space associated with the non-canonical images and a position and rotation associated with the canonical image. Training updates the orientation neural network with a loss function configured to reduce a distance between a calculated transformation produced by the orientation neural network in training and a ground truth transformation for each non-canonical image. In some embodiments, a training converter produces training transformations using canonical image together with non-canonical images and conversions associated with positions and rotations derived from inertial measurement unit data.
Additional embodiments include a diagnoser configured to make a diagnosis from a final image generated by the probe when at a position and rotation similar to those associated with a canonical image, and further processing to check sufficiency and reconstruct a 3D/4D volume by applying predicted transformations to rotate cross-sections followed by tomographic reconstruction. Other embodiments include generating sets of cyclical canonical images during a body part cycle using a trained cyclical canonical view neural network, with a set creator generating sets of pairs of new images and generated transformations, and a sufficiency checker determining when enough sets have been created.
Claims Coverage
The provided excerpt contains three independent claims: clm-00001, clm-00004, clm-00008, and clm-00011. Across these independent claims, the core coverage is a trained orientation neural network that maps a non-canonical image to a transformation between canonical and non-canonical poses, plus conversion of that transformation into probe position and/or rotation instructions for user guidance, with separate provisions for training using a loss function based on ground truth transformations and additional components such as set generation, sufficiency checking, and generation of cyclical canonical images.
Orientation neural network generating canonical-to-non-canonical pose transformation
A trained orientation neural network to receive a non-canonical image of a body part from said ultrasound unit and to generate a transformation associated with said non-canonical image to transform between a position and rotation associated with a canonical image and a position and rotation associated with said non-canonical image.
Result converter converting transformation into probe position/rotation instructions
A result converter to convert said transformation into position and/or rotation instructions of said probe and to display said position and/or rotation instructions to a user to change the position and/or rotation of said probe.
Training orientation neural network using canonical and non-canonical images with loss on transformation distance
A trainer to train an orientation neural network in training using said canonical image together with non-canonical images taken around said canonical image and transformations between positions and rotations in space associated with said non-canonical images and a position and rotation associated with said canonical image, wherein said trainer is configured to produce said trained orientation neural network, wherein said trainer is configured to update said orientation neural network in training with a loss function, wherein said loss function is configured to reduce a distance between a calculated transformation produced by said orientation neural network in training and a ground truth transformation for each non-canonical image.
Set creation and sufficiency checking for training pairs
A set creator to receive a multiplicity of new images from said probe and a multiplicity of new transformations from said trained orientation neural network in response to said multiplicity of new images and to generate therefrom sets of pairs, each pair comprising one of said multiplicity of new images and its generated transformation; a sufficiency checker to determine when enough sets have been created.
Cyclical canonical view neural network generating cyclical canonical images
A trained cyclical canonical view neural network to generate a set of cyclical canonical images showing changes in said body part during a body part cycle.
Method receiving non-canonical image and generating transformation using trained orientation neural network
Receiving a non-canonical image of a body part from said ultrasound unit and generating, using a trained orientation neural network, a transformation associated with said non-canonical image and transforming between a position and rotation associated with a canonical image and a position and rotation associated with said non-canonical image.
Method converting transformation into probe position/rotation instructions and displaying to user
Converting said transformation into position and/or rotation instructions of said probe and displaying said position and/or rotation instructions to a user to change the position and/or rotation of said probe.
Method training orientation neural network with loss function reducing distance to ground truth transformations
Before said receiving, training an orientation neural network in training using said canonical image together with non-canonical images taken around said canonical image and transformations between positions and rotations in space associated with said non-canonical images and a position and rotation associated with said canonical image, said training producing said trained orientation neural network, said training comprising: utilizing a loss function to update said orientation neural network in training, wherein said loss function is configured to reduce a distance between a calculated transformation produced by said orientation neural network in training and a ground truth transformation for each non-canonical image.
Method generating sets of image/transformation pairs and generating cyclical canonical images
Receiving a multiplicity of new images from said probe and a multiplicity of transformations from said trained orientation neural network in response to said multiplicity of new images and generating therefrom sets of pairs, each pair comprising one of said multiplicity of images and its generated transformation; determining when enough sets have been created; and generating, using a trained cyclical canonical view neural network, a set of cyclical canonical images showing changes in said body part during a body part cycle.
The independent claims focus on a trained orientation neural network that outputs a transformation between canonical and non-canonical pose (position and rotation), a result converter that turns that transformation into probe position and/or rotation instructions displayed to the user, and training updated using a loss function that reduces distance to ground truth transformations. Claim clm-00004 and clm-00011 additionally cover set creation with a sufficiency checker and generating cyclical canonical images using a trained cyclical canonical view neural network.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?