Method and device for automatic motion detection in medical image-series
In a method for automatic motion detection in medical image-series, a dataset of a series of images is provided. The images can be of a similar region of interest that are recorded at consecutive points of time. The method can further include localizing a target in the images of the dataset and calculating a position of the target in the images to calculate localization data of the target, and calculating movement data of a movement of the target of temporal adjacent images of the images based on the localization data.
1. A method for automatic motion detection in medical image-series, comprising:
providing a dataset of a series of images, the images being of a similar region of interest that are recorded at consecutive points of time;
localizing, using a localizer that is a first deep neural network, a target in the images of the dataset and calculating a position of the target in the images to calculate localization data of the target; and
calculating, using a motion detector that is a second deep neural network, movement data of a movement of the target of temporal adjacent images of the images based on the localization data, wherein the first deep neural network of the localizer and the second deep neural network of the motion detector are formed as chained deep neural networks.
2. The method according to claim 1 , wherein calculating the localization data comprises performing calculations based on: (a) probabilistic boosting trees and/or random forests or (b) an adaptive algorithm trained for localizing the target in the images of the dataset, and calculating the position of the target in the images.
3. The method according to claim 1 , wherein the movement data is calculated based on: (a) image registration of the target in the temporal adjacent images or (b) an adaptive algorithm trained for calculating movement data of the temporal adjacent images based on the localization data, wherein the calculation of the movement data of the target and the calculation of the localization data is performed in a convoluted process with a same adaptive algorithm.
4. The method according to claim 1 , wherein calculating the movement data of the movement of the target generates a quantitative motion curve describing an average motion magnitude between the temporal adjacent images, the calculating the movement data including:
performing an elastic image registration on the series of images combined with averaging lengths of motion vectors between the temporal adjacent images, or
training an adaptive algorithm with the elastic image registration on the series of images combined with averaging the lengths of motion vectors between the temporal adjacent images.
5. The method according to claim 1 , further comprising comparing the calculated movement data of the target with a predefined threshold movement value to perform a motion detection.
6. The method according to claim 5 , wherein the threshold movement value is obtained from one or more predefined annotated example cases, selected based on physiological knowledge, and/or selected based on a predefined image resolution.
7. The method according to claim 1 , wherein, before calculating the movement data, the localization data is used for cropping areas as a bounding box from the images containing the target, wherein, in case of segmentation, a tightest fitting bounding box enclosing contours of all timepoints of the target is selected.
8. The method according to claim 1 , wherein the method provides an automatic motion detection in CINE data, the target in the images including image data of coronary arteries, right or left atrium, and/or right or left ventricle.
9. A system for automatic motion detection in medical image-series, the system comprising:
a data interface configured to receive a dataset of a series of images of a similar region of interest that are recorded at consecutive points of time;
a localizer that is a first deep neural network and configured to localize a target in the images of the dataset to calculate localization data of the target, and calculate a position of the target in the images; and
a motion detector that is a second deep neural network and configured to calculate movement data of a movement of the target of temporal adjacent images of the images based on the localization data, wherein the first deep neural network of the and the second deep neural network of the motion detector are formed as chained deep neural networks.
10. The system according to claim 9 , wherein the localizer comprises:
a landmark detector configured to obtain landmark position data and/or a center point based on a position of the target; and/or
a segmentor configured to obtain contours around the target or a segmented region.
11. The system according to claim 10 , wherein the landmark detector comprises a 3D-DenseNet, and the segmentor comprises a U-net structure.
12. The system according to claim 9 , wherein the localizer comprises:
a landmark detector configured to obtain landmark position data and/or a center point based on a position of the target; and
a segmentor configured to obtain contours around the target or a segmented region.
13. The system according to claim 9 , wherein the localizer and/or the motion detector comprise a trained machine learning network.
14. The system according to claim 13 , wherein the trained machine learning network comprises a neural network, a convolutional network, and/or an end-to-end network.
15. The system according to claim 9 , wherein the localizer and the motion detector are part of a data-network in data-communication with a magnetic resonance imaging system, wherein the data network includes parts of the internet and/or a cloud-based computing system.
16. A controller configured to control a medical imaging system comprising the system according to claim 9 .
17. A medical imaging system comprising:
a medical image scanner; and
a controller comprising a localizer that is a first deep neural network and a motion detector that is a second deep network, the controller being configured to:
generate a dataset of a series of images, the images being of a similar region of interest that are recorded at consecutive points of time;
localize, using the first deep neural network of the localizer, a target in the images of the dataset and calculating a position of the target in the images to calculate localization data of the target; and
calculate, using the second deep neural network of the motion detector, movement data of a movement of the target of temporal adjacent images of the images based on the localization data, wherein the first deep neural network of the localizer and the second deep neural network of the motion detector are formed as chained deep neural networks.
18. A non-transitory computer program product having a computer program which is directly loadable into a memory of a medical imaging system, when executed a controller of the system, causes the system to perform the method as claimed in claim 1 .
19. A non-transitory computer-readable storage medium with an executable program stored thereon, that when executed, instructs a processor to perform the method of claim 1 .
20. The method according to claim 1 , wherein the output of the first deep neural network is provided as an input to the second deep neural network to form the chained deep neural networks.