Granted Patent B2

US 12,217,434 · App. 18/208,331 · Granted Feb 4, 2025

Inventor: Marina Yaacobi (Beer-Sheva, IL)

Assignee: KONINKLIJKE PHILIPS N.V.

G06T7/11G06T3/4007G06T2207/10132G06T2207/20081G06T2207/20084G06T2207/20104G06T2207/20112G06T2207/30048

Abstract

Claims (62)

1. A computer-implemented method of automatically processing two-dimensional (2D) ultrasound images for computing of at least one clinical parameter of a right ventricle (RV), comprising:

selecting one 2D ultrasound image of a plurality of 2D ultrasound images depicting at least a RV of a subject, sequentially captured over at least one cardiac cycle of the subject;

classifying the RV depicted in the selected one 2D ultrasound image into a predefined shape selected from a plurality of predefined shapes;

interpolating an inner contour of an endocardial border of the RV for the selected one 2D ultrasound image according to the classified predefined shape of the selected one 2D ultrasound image;

tracking the interpolated inner contour obtained for the one 2D ultrasound image over the plurality of 2D images over the at least one cardiac cycle;

computing, a RV area of the RV for each respective 2D ultrasound image of the plurality of 2D ultrasound images, according to the tracked interpolated inner contour;

identifying a first 2D ultrasound image depicting an end-diastole (ED) state according to a maximal value of the RV area for the plurality of 2D images, and a second 2D US image depicting an end-systole(ES) state according to minimal value of the RV area for the plurality of 2D images; and

computing at least one clinical parameter of the RV according to the identified first 2D ultrasound image depicting the ED state and the second 2D US image depicting the ES state.

2. The method of claim 1 , further comprising, wherein the at least one clinical parameter is computed according to the first 2D ultrasound image depicting the ED state and the second 2D US image depicting the ES state with interpolated inner contour.

3. The method of claim 2 , further comprising:

inputting the selected 2D image into a trained machine learning (ML) model trained on a training dataset including for each of a plurality of sample subjects, at least one 2D ultrasound image depicting a RV and a corresponding plurality of segmentation masks including: a RV cavity segmentation mask, a lateral tricuspid valve annulus segmentation mask, and a septal tricuspid valve annulus segmentation mask; and

obtaining, for the selected 2D image, as an outcome of the ML model, a plurality of segmentation masks including: a RV cavity segmentation mask, a lateral tricuspid valve annulus segmentation mask, and a septal tricuspid valve annulus segmentation mask,

extracting, an inner contour of an endocardial border of the RV for the selected one 2D ultrasound image based on the plurality of segmentation masks,

extracting a plurality of valve indications for the selected one 2D ultrasound image,

wherein the plurality of valve indications include a plurality of positions of a tricuspid valve extracted from a center region of the lateral tricuspid valve annulus segmentation mask, and extracted from a center region of the septal tricuspid valve annulus segmentation mask.

4. The method of claim 3 , further comprising automatically selecting a region of interest from the selected 2D image, the ROI including the RV cavity, the lateral tricuspid valve annulus, and the septal tricuspid valve annulus, wherein inputting the selected 2D image comprises inputting the ROI into the trained ML model, wherein the at least one 2D ultrasound image of the training dataset comprises the ROI image created from the at least one 2D ultrasound image of the training dataset.

5. The method of claim 3 , further comprising, for the selected 2D image of the plurality of 2D ultrasound images:

extracting a respective border of the RV depicted in the selected 2D image using the respective cavity mask; and

generating a binary image using the respective extracted RV border,

wherein the extracting of the inner contour for the selected 2D image is based on the binary image.

6. The method of claim 5 , further comprising, for each 2D image of the plurality of 2D ultrasound images:

segmenting around the respective extracted RV border on sampled points using local contrast enhancement, wherein the binary image is of the respective extracted RV border with segmentation.

7. The method of claim 1 , further comprising:

identifying a tricuspid valve of the inner contour;

identifying an apex of the RV on the inner contour,

dividing the inner contour into a lateral side and a septal side with respect to the apex and the tricuspid valve,

wherein interpolating the inner contour is done each of the lateral side and the septal side according to the classified shape.

8. The method of claim 7 , wherein the classified shape is selected from one of “triangle” and “square”, and the lateral side is interpolated using a first interpolation method when the shape is triangle, and the lateral side is interpolated using a second interpolation method when the shape is square.

9. The method of claim 1 , wherein the 2D ultrasound images exclude a correlation with electrocardiogram (ECG) data of the subject, and the ED state and the ES state are identified from the plurality of 2D ultrasound images without the correlation with ECG data.

10. The method of claim 1 , further comprising:

creating a model of a trajectory of a tricuspid annulus location on a free wall between each pair of ED and ES denoting systole, and ES and next ED 2D images denoting diastole;

comparing a predicted valve location to actual location from tracking and correct high deviation using the model and repeat interpolation to compute a deviation;

when the deviation is above a threshold, the tracked valve location is corrected according to the model, and a border point interpolation process is repeated taking into account a corrected location of a left annulus point to obtain a corrected inner contour;

calculate area of a chamber denoting the area enclosed in the tracked contours for each 2D ultrasound image based on the corrected inner contour;

wherein the at least one clinical parameter comprises the calculated area.

11. The computer implemented method of claim 1 , wherein tracking further comprises tracking a plurality of locations of a tricuspid valve obtained for the one 2D ultrasound image, and computing the RV area further comprises computing the RV area according to the tracked plurality of locations of the tricuspid valve.

12. The computer implemented method of claim 1 , wherein the at least one clinical parameter comprises at least one of: (i) a fractional area change (FAC) for the RV, (ii) a free wall strain (FWS) for the RV denoting the maximum strain of the free wall, wherein the FWS is computed for each respective cardiac cycle for the RV, RV basal diameter (RVD1), RV mid diameter (RVD2) and RV length (RVD3).

13. The computer implemented method of claim 1 , wherein the at least one clinical parameter comprises at least one of: (i) a tricuspid annular plane systolic excursion (TAPSE), and further comprising computing the TAPSE by:

computing the distance of lateral free wall tricuspid valve annulus from an apex in ED to obtain a first value,

computing the distance of lateral free wall tricuspid valve annulus from the apex in ES to obtain a second value, and computing a difference between the first value and the second value, and

(ii) a peak systolic velocity of a tricuspid annulus (S′), and further comprising computing the S′ by: computing a displacement curve by computing a respective distance of lateral free wall tricuspid valve annulus from an apex each of the plurality of 2D US images of a respective cardiac cycle and subtracting from each respective distance the relative displacement denoting a change in distance, wherein the change in distance is computed by deriving a distance curve of the respective 2D image with respect to frame time to obtain the velocity of movement in a corresponding location, wherein S′ denotes the maximum of the velocity value per respective cardiac cycle.

14. The method of claim 1 , wherein each of the at least one clinical parameters is computed for each cardiac cycle depicted in the plurality of 2D ultrasound images.

15. The method of claim 1 , further comprising treating the subject for a medical condition selected from the group consisting of: impaired heart function, cardiac injury due to COVID-19, pulmonary embolism, pulmonary hypertension (PH), congenital heart disease (CHD), coronary artery disease, left-sided heart failure (HF), right-sided heart failure, and valvular heart disease, using a treatment effective for the medical condition, selected from the group consisting of: fibrinolysis, thrombolytic, anticoagulant, pacemaker, beta blockers, nitrates, oxygen, morphine, ACE inhibitors, insulin, ATII receptor blockers (ARBs), diuretics, cardiac defibrillation, cardiac surgery, trans catheterization, stent placement, valve replacement.

16. The method of claim 1 , further comprising iteratively computing the at least one clinical parameter over a plurality of time intervals, and plotting, on a display, a trend line of the at least one clinical parameters for the plurality of time intervals.

17. The method of claim 1 , further comprising selecting a computational processes from a plurality of computational process according to the predefined shape selected from the plurality of predefined shapes.

18. A system for automatically processing two dimensional (2D) ultrasound images for computing of at least one clinical parameter of a right ventricle (RV), comprising:

at least one hardware processor executing a code for:

selecting one 2D ultrasound image of a plurality of 2D ultrasound images depicting at least a RV of a subject, sequentially captured over at least one cardiac cycle of the subject;

classifying the RV depicted in the selected one 2D ultrasound image into a predefined shape selected from a plurality of predefined shapes;

interpolating an inner contour of an endocardial border of the RV for the selected one 2D ultrasound image according to the classified predefined shape of the selected one 2D ultrasound image;

tracking the interpolated inner contour obtained for the one 2D ultrasound image over the plurality of 2D images over the at least one cardiac cycle;

computing, a RV area of the RV for each respective 2D ultrasound image of the plurality of 2D ultrasound images, according to the tracked interpolated inner contour;

identifying a first 2D ultrasound image depicting an end-diastole (ED) state according to a maximal value of the RV area for the plurality of 2D images, and a second 2D US image depicting an end-systole(ES) state according to minimal value of the RV area for the plurality of 2D images; and

computing at least one clinical parameter of the RV according to the identified first 2D ultrasound image depicting the ED state and the second 2D US image depicting the ES state.

19. A non-transitory computer readable medium for automatically processing two dimensional (2D) ultrasound images for computing of at least one clinical parameter indicative of a right ventricle (RV) comprising program instructions which, when executed by a processor, cause the processor to perform:

selecting one 2D ultrasound image of a plurality of 2D ultrasound images depicting at least a RV of a subject, sequentially captured over at least one cardiac cycle of the subject;

classifying the RV depicted in the selected one 2D ultrasound image into a predefined shape selected from a plurality of predefined shapes;

interpolating an inner contour of an endocardial border of the RV for the selected one 2D ultrasound image according to the classified predefined shape of the selected one 2D ultrasound image;

tracking the interpolated inner contour obtained for the one 2D ultrasound image over the plurality of 2D images over at least one cardiac cycle;

computing, a RV area of the RV for each respective 2D ultrasound image of the plurality of 2D ultrasound images, according to the tracked interpolated inner contour;

identifying a first 2D ultrasound image depicting an end-diastole (ED) state according to a maximal value of the RV area for the plurality of 2D images, and a second 2D US image depicting an end-systole(ES) state according to minimal value of the RV area for the plurality of 2D images; and

computing at least one clinical parameter of the RV according to the identified first 2D ultrasound image depicting the ED state and the second 2D US image depicting the ES state.

Assignments (2)

Continuity (2)

References Cited (24)