Intramyocardial tissue displacement and motion measurement and strain analysis from MRI cine images using DENSE deep learning
An exemplary method and system are disclosed that employ DENSE deep learning neural-network(s) trained with displacement-encoded imaging data (i.e., DENSE data) to estimate intramyocardial motion from cine MRI images and other cardiac medical imaging modalities, including standard cardiac computer tomography (CT) images, magnetic resonance imaging (MRI) images, echocardiogram images, heart ultrasound images, among other medical imaging modalities described herein. The DENSE deep learning neural-network(s) can be configured (trained) using (i) contour motion data from displacement-encoded imaging magnitude data as inputs to the neural network and (ii) displacement maps derived from displacement-encoded imaging phase images for comparison to the outputs of the neural network for neural network adjustments during the training.
1 . A method of determining intramyocardial motion and/or measurand in medical image scans, the method comprising:
retrieving, by a processor, a medical image scan of a subject;
determining, by the processor, intramyocardial motion data in the medical image scan, in part, using a trained neural network, wherein the trained neural network has been trained by:
(i) generating a set of contour motion images or data from Displacement-ENcoding with Stimulated Echoes (DENSE) magnitude images or data and a set of training displacement map images or data from DENSE phase images or data; and
(ii) configuring a neural network comprising an input and output to generate the trained neural network by (a) applying the set of contour motion images or data to the input to generate an output displacement map image or data and (b) applying the set of the training displacement map images or data, or parameters derived therefrom, to the output displacement map to adjust weights of the neural network;
wherein the determined intramyocardial motion data, or a parameter derived therefrom, is outputted in a report or employed in a control operation for the diagnostics or treatment of cardiac disease or cardiac health-related conditions.
2 . The method of claim 1 , wherein the medical image scan is at least one of: cardiac computer tomography (CT) images, magnetic resonance imaging (MRI) images, echocardiogram images, and heart ultrasound images.
3 . The method of claim 1 further comprising:
determining, by the processor, a set of values associated with at least one of a strain parameter, a strain rate parameter, a torsion parameter, and a twist parameter using the determined intramyocardial motion data, wherein the determined set of values associated with the at least one of the strain parameter, the strain rate parameter, the torsion parameter, and the twist parameter is outputted in the report or employed in the control operation for the diagnostics or the treatment of cardiac disease or the cardiac health-related conditions.
4 . The method of claim 1 , wherein the set of contour motion images or data is generated by:
binarizing pixels of the DENSE magnitude images or data to a binary value corresponding to contour motions defined within the DENSE magnitude images or data.
5 . The method of claim 4 , wherein the set of contour motion images or data is further generated by:
scaling the DENSE magnitude images or data to a pre-defined image size; and
cropping the scaled DENSE magnitude images or data to an image region of interest.
6 . The method of claim 1 , wherein the neural network comprises a convolutional neural network comprising one or more convolution layers, one or more batch normalization layers, one or more ReLU layers, and one or more pooling layers, the layers being connected together collectively to form a network.
7 . The method of claim 1 , wherein the neural network comprises a 3D UNet neural network.
8 . The method of claim 1 , wherein the DENSE magnitude images or data and the DENSE phase images or data are determined from a plurality of DENSE training data sets, wherein the plurality of DENSE training data sets are acquired by:
acquiring first data comprising a stimulated echo and a Tl relaxation echo;
acquiring second data comprised of a second stimulated echo, a second Tl relaxation echo, and a second stimulated anti-echo;
acquiring at least one of original frames comprising the stimulated echo and the Tl relaxation echo;
acquiring at least one of additional original frames comprising a stimulated echo, a Tl relaxation echo, and a stimulated anti-echo; and
acquiring a plurality of new original frames of displacement encoded stimulated echo (DENSE) cine frames of MRI image data of a subject.
9 . A method of training a neural network having an input and output to generate an output displacement map corresponding to intramyocardial motion in a biomedical image, the method comprising:
generating, via a processor, a set of contour motion images or data from DENSE magnitude images or data;
generating, via the processor, a set of training displacement map images or data from DENSE phase images or data; and
configuring the neural network by (i) applying the set of contour motion images or data to the input of the neural network to generate an output displacement map image or data and (ii) applying the set of training displacement map images or data, or parameter derived therefrom, to the output displacement map to adjust weights of the neural network,
wherein the trained neural network is employed to determine intramyocardial motion data, or a parameter derived therefrom, and wherein output of the trained neural network is outputted in a report or employed in a control operation for the diagnostics or treatment of cardiac disease or cardiac health-related conditions.
10 . The method of claim 9 , wherein the medical image scan is at least one of: cardiac computer tomography (CT) images, magnetic resonance imaging (MRI) images, echocardiogram images, heart ultrasound images.
11 . The method of claim 9 , wherein the determined intramyocardial motion data is employed to determine a set of values associated with at least one of a strain parameter, a strain rate parameter, a torsion parameter, and a twist parameter, and wherein the determined set of values associated with the at least one of the strain parameter, the strain rate parameter, the torsion parameter, and the twist parameter is outputted in the report or employed in the control operation for the diagnostics or the treatment of cardiac disease or the cardiac health-related conditions.
12 . The method of claim 9 , wherein the set of contour motion images or data is generated by:
scaling the DENSE magnitude images or data to a pre-defined image size;
cropping the scaled DENSE magnitude images or data to an image region of interest; and
binarizing pixels of the DENSE magnitude images or data to a binary value corresponding to contour motions defined within the DENSE magnitude images or data.
13 . A system comprising:
a processor; and
a memory having instructions stored thereon to determine intramyocardial motion and/or measurand in medical image scans, wherein execution of the instructions by the processor causes the processor to:
retrieve medical image scan of a subject;
determine intramyocardial motion data in the medical image scan, in part, using a trained neural network, wherein the trained neural network has been trained by:
(i) generating a set of contour motion images or data from Displacement-ENcoding with Stimulated Echoes (DENSE) magnitude images or data and a set of training displacement map images or data from DENSE phase images or data; and
(ii) configuring a neural network comprising an input and output to generate the trained neural network by (a) applying the set of contour motion images or data to the input to generate an output displacement map image or data and (b) applying the set of training displacement map images or data to the output displacement map to adjust weights of the neural network;
wherein the determined intramyocardial motion data, or a parameter derived therefrom, is outputted in a report or employed in a control operation for the diagnostics or treatment of cardiac disease or cardiac health-related conditions.
14 . The system of claim 13 , wherein the medical image scan is at least one of: cardiac computer tomography (CT) images, magnetic resonance imaging (MRI) images, echocardiogram images, and heart ultrasound images.
15 . The system of claim 13 , wherein the determined intramyocardial motion data is employed to determine a set of values associated with at least one of a strain parameter, a strain rate parameter, a torsion parameter, and a twist parameter, and wherein the determined set of values associated with the at least one of the strain parameter, the strain rate parameter, the torsion parameter, and the twist parameter is outputted in the report or employed in the control operation for the diagnostics or the treatment of cardiac disease or the cardiac health-related conditions.
16 . The system of claim 13 , wherein the instructions include:
a first instruction to scale the DENSE magnitude images or data to a pre-defined image size;
a second instruction to crop the scaled DENSE magnitude images or data to an image region of interest; and
a third instruction to binarize pixels of the DENSE magnitude images or data to a binary value corresponding to contour motions defined within the DENSE magnitude images or data.
17 . A non-transitory computer-readable medium having instructions stored thereon to determine intramyocardial motion and/or measurand in medical image scans, wherein execution of the instructions by the processor causes the processor to:
retrieve medical image scan of a subject;
determine intramyocardial motion data in the medical image scan, in part, using a trained neural network, wherein the trained neural network has been trained by:
(i) generating a set of contour motion images or data from Displacement-ENcoding with Stimulated Echoes (DENSE) magnitude images or data and a set of training displacement map images or data from DENSE phase images or data; and
(ii) configuring a neural network comprising an input and output to generate the trained neural network by (a) applying the set of contour motion images or data to the input to generate an output displacement map image or data and (b) applying the set of training displacement map images or data to the output displacement map to adjust weights of the neural network;
wherein the determined intramyocardial motion data, or a parameter derived therefrom, is outputted in a report or employed in a control operation for the diagnostics or treatment of cardiac disease or cardiac health-related conditions.
18 . The non-transitory computer-readable medium of claim 17 , wherein the medical image scan is at least one of: cardiac computer tomography (CT) images, magnetic resonance imaging (MRI) images, echocardiogram images, and heart ultrasound images.
19 . The non-transitory computer-readable medium of claim 17 , wherein the determined intramyocardial motion data is employed to determine a set of values associated with at least one of a strain parameter, a strain rate parameter, a torsion parameter, and a twist parameter, and wherein the determined set of values associated with the at least one of the strain parameter, the strain rate parameter, the torsion parameter, and the twist parameter is outputted in the report or employed in the control operation for the diagnostics or the treatment of cardiac disease or the cardiac health-related conditions.
20 . The non-transitory computer-readable medium of claim 17 , wherein the instructions include:
a first instruction to scale the DENSE magnitude images or data to a pre-defined image size;
a second instruction to crop the scaled DENSE magnitude images or data to an image region of interest; and
a third instruction to binarize pixels of the DENSE magnitude images or data to a binary value corresponding to contour motions defined within the DENSE magnitude images or data.