Automated measurement based on deep learning
A framework for automated measurement. In accordance with one aspect, the framework detects a centerline point of a structure of interest in an image. A centerline of the structure of interest may be traced based on the detected centerline point. A trained segmentation learning structure may be used to generate one or more contours of the structure of interest along the centerline. One or more measurements may then be extracted from the one or more contours.
1. A system for automated measurement, comprising:
a non-transitory memory device for storing computer readable program code; and
a processor device in communication with the memory device, the processor being operative with the computer readable program code to perform steps including
receiving an image of an aorta,
detecting a centerline point of the aorta in the image,
tracing a centerline of the aorta using the detected centerline point as a starting point,
generating, by a trained segmentation learning structure, one or more contours of cross-sections of the aorta along the centerline, wherein the trained segmentation learning structure comprises a deep learning architecture trained by two-dimensional cross-sectional images of the aorta,
extracting one or more measurements from the one or more contours, and
presenting the one or more measurements in a graphical user interface.
2. The system of claim 1 wherein the processor is operative with the computer readable program code to extract multiple measurements continuously along the centerline of the aorta and present the multiple measurements as a continuous profile along the aorta in the graphical user interface.
3. The system of claim 1 wherein the processor is operative with the computer readable program code to extract multiple measurements at clinically standard points along the centerline of the aorta.
4. The system of claim 1 wherein the processor is operative with the computer readable program code to generate the one or more contours of the aorta along the centerline by
extracting, from the received image, a cross-sectional image of the aorta, and
providing the cross-sectional image to the trained segmentation learning structure to generate the one or more contours.
5. The system of claim 1 wherein the processor is operative with the computer readable program code to present the one or more measurements in the graphical user interface by overlaying the one or more measurements with corresponding population measurements in a measurement panel.
6. A method of automated measurement, comprising:
receiving an image of a structure of interest;
tracing a centerline of the structure of interest in the image;
generating, by a trained segmentation learning structure, one or more contours of cross-sections of the structure of interest along the centerline, wherein the trained segmentation learning structure comprises a deep learning architecture trained by two-dimensional cross-sectional images of the structure of interest;
extracting one or more measurements from the one or more contours; and
presenting the one or more measurements in a graphical user interface.
7. The method of claim 6 wherein tracing the centerline of the structure of interest comprises:
detecting a centerline point of the structure of interest; and
tracing the centerline of the structure of interest using the centerline point as a starting point.
8. The method of claim 7 wherein tracing the centerline comprises using a regression forest-based orientation detector that is trained to trace the centerline.
9. The method of claim 7 wherein tracing the centerline of the structure of interest comprises repeatedly detecting a direction of the centerline at a current point and stepping to a next point along the detected direction.
10. The method of claim 6 wherein generating the one or more contours of the structure of interest along the centerline comprises:
extracting, from the received image, a cross-sectional image of the structure of interest; and
providing the cross-sectional image to the trained segmentation learning structure to generate the one or more contours.
11. The method of claim 10 further comprises training the segmentation learning structure using a set of two-dimensional images of different aortas from different patients.
12. The method of claim 10 wherein the trained segmentation learning structure comprises a convolutional neural network classifier.
13. The method of claim 6 wherein extracting the one or more measurements from the one or more contours comprises extracting multiple measurements continuously along the centerline of the structure of interest, and wherein presenting the one or more measurements comprises presenting the multiple measurements as a continuous profile along the structure of interest in the graphical user interface.
14. The method of claim 6 wherein extracting the one or more measurements from the one or more contours comprises extracting a maximum diameter of at least one of the one or more contours.
15. The method of claim 6 wherein extracting the one or more measurements from the one or more contours comprises extracting a diameter perpendicular to an axis along a maximum diameter of at least one of the one or more contours.
16. The method of claim 6 wherein extracting the one or more measurements from the one or more contours comprises extracting an average diameter of the one or more contours.
17. The method of claim 6 wherein extracting the one or more measurements from the one or more contours comprises extracting an area defined by at least one of the one or more contours.
18. The method of claim 6 wherein presenting the one or more measurements in the graphical user interface comprises overlaying the one or more measurements with corresponding population measurements in a measurement panel to detect any abnormal dilation of the structure of interest.
19. The method of claim 6 wherein presenting the one or more measurements in the graphical user interface comprises presenting the one or more measurements in a measurement panel and linking the measurement panel with an image visualization panel.
20. One or more non-transitory computer-readable media embodying instructions executable by machine to perform operations, comprising:
receiving an image of a structure of interest;
tracing a centerline of the structure of interest in the image;
generating, by a trained segmentation learning structure, one or more contours of cross-sections of the structure of interest along the centerline, wherein the trained segmentation learning structure comprises a deep learning architecture trained by two-dimensional cross-sectional images of the structure of interest;
extracting one or more measurements from the one or more contours; and
presenting the one or more measurements in a graphical user interface.