Systems and methods for automatically detecting anatomical features for preoperative cardiac implant simulations
Systems and methods for fully automated anatomical analysis of an anatomical structure are provided to facilitate pre-operative planning. The computerized method may include obtaining a plurality of images, e.g., MSCT images, of patient-specific cardiovascular anatomy, and analyzing the MSCT images with a trained artificial intelligence module to identify one or more anatomical landmarks and to construct a virtual three-dimensional model of the anatomical structure. For example, the trained artificial intelligence module may execute segmentation, point detection, curve detection, or plane detection deep learning modules, independently or in combination, to identify the anatomical landmarks. The method further may include deriving anatomical measurements of the one or more identified anatomical landmarks, and displaying the virtual three-dimensional model alongside the anatomical measurements of the one or more identified anatomical landmarks.
1 . A computerized method for automated anatomical analysis of an anatomical structure, the method comprising:
obtaining a plurality of images of patient-specific cardiovascular anatomy, said plurality of images comprising a plurality of voxels;
analyzing the plurality of images with a trained artificial intelligence module to identify one or more anatomical landmarks and to construct a virtual three-dimensional model of the anatomical structure, wherein analyzing the plurality of images with the trained artificial intelligence module comprises:
generating a probability mask, the probability mask comprising a probability value for each voxel of the plurality of voxels, the probability value indicating a probability that said voxel corresponds to the anatomical structure;
converting the probability mask to a segmentation mask by comparing each probability value to a threshold value on a voxel-by-voxel basis to assign each voxel of the plurality of voxels a predetermined label corresponding to the anatomical structure if such probability value exceeds the threshold value; and
generating a segmentation mask comprising each voxel of the plurality of voxels assigned the predetermined label; and
constructing the virtual three-dimensional model of the anatomical structure based on the segmentation mask;
deriving anatomical measurements based on the one or more identified anatomical landmarks;
obtaining specifications of a plurality of implantable devices associated with the anatomical structure;
displaying a sizing chart based on the specifications of the plurality of implantable devices and on the anatomical measurements;
receiving input from a user comprising a selection of an implantable device from the plurality of implantable devices; and
displaying the virtual three-dimensional model of the anatomical structure alongside the anatomical measurements of the one or more identified anatomical landmarks and a virtual model of the selected implantable device.
2 . The computerized method of claim 1 , wherein the anatomical structure is a left atrium and left atrial appendage.
3 . The computerized method of claim 2 , wherein the one or more anatomical landmarks comprise an ostium and a predetermined landing zone within the anatomical structure for a cardiac implantable device.
4 . The computerized method of claim 3 , wherein deriving anatomical measurements based on the one or more identified anatomical landmarks comprises identifying 3D planes defining the ostium and the predetermined landing zone, and performing measurements in the 3D planes.
5 . The computerized method of claim 3 , wherein deriving anatomical measurements based on the one or more identified anatomical landmarks comprises measuring a depth of the left atrial appendage.
6 . The computerized method of claim 2 , wherein the one or more anatomical landmarks further comprise a fossa ovalis, such that the fossa ovalis is identified as a 3D curve on an interatrial septum via the trained artificial intelligence module.
7 . The computerized method of claim 6 , further comprising planning a transseptal puncture site based on the identified fossa ovalis.
8 . The computerized method of claim 2 , wherein the one or more anatomical landmarks further comprise a mitral valve annulus.
9 . The computerized method of claim 1 , wherein the anatomical structure is an aortic root.
10 . The computerized method of claim 9 , wherein the one or more anatomical landmarks comprise an aortic annulus, a left ventricular outflow tract, a sino-tubular junction, or a sinus of Valsalva.
11 . The computerized method of claim 10 , wherein deriving anatomical measurements based on the one or more identified anatomical landmarks comprises measuring at least one of left coronary distance, right coronary distance, sino-tubular junction distance, aortic arch angulation, or membranous septum length.
12 . The computerized method of claim 1 , wherein obtaining the plurality of images of patient-specific cardiovascular anatomy comprises obtaining a plurality of multi-slice computed tomography images of patient-specific cardiovascular anatomy.
13 . The computerized method of claim 12 , further comprising pre-processing the plurality of multi-slice computed tomography images by resampling a volume of the multi-slice computed tomography images to an isotropic resolution and voxel size.
14 . The computerized method of claim 1 , wherein analyzing the plurality of images with the trained artificial intelligence module to identify one or more anatomical landmarks comprises executing at least one deep learning module selected from a list consisting of: segmentation, point detection, curve detection, and plane detection.
15 . The computerized method of claim 1 , wherein assigning the predetermined label to each voxel of the plurality of voxels if the probability value exceeds the threshold value comprises binarizing the probability mask for a predefined class such that each voxel with a probability value below the threshold value is set to label zero and each voxel with a probability value above the threshold value is set to label one, and wherein the segmentation mask comprises each voxel having label one.
16 . The computerized method of claim 1 , further comprising combining the segmentation mask with one or more second segmentation masks obtained using an image analysis technique.
17 . The computerized method of claim 16 , wherein the image analysis technique is flooding.
18 . The computerized method of claim 14 , wherein executing the point detection deep learning module comprises:
generating a second probability mask indicative of a second probability value that each voxel of a plurality of voxels of the plurality of images is a predetermined label defined by a predefined spherical region around a predefined point;
assigning the predetermined label to each voxel if the second probability value exceeds a second threshold; and
obtaining a 3D point by taking a centroid of all voxels having the predetermined label to identify the one or more anatomical landmarks.
19 . The computerized method of claim 18 , wherein identifying the one or more anatomical landmarks further comprises cropping the plurality of images around the 3D point.
20 . The computerized method of claim 14 , wherein executing the curve detection deep learning module comprises:
generating a second probability mask indicative of a second probability value that each voxel of a plurality of voxels of the plurality of images is a predetermined label defined by a curve formed by sweeping a sphere having a predetermined radius along the curve for one of the one or more anatomical landmarks;
assigning the predetermined label to each voxel if the second probability value exceeds a second threshold; and
identifying a 3D curve by using a graph-based technique on all voxels having the predetermined label.
21 . The computerized method of claim 14 , wherein executing the plane detection deep learning module comprises:
assigning each voxel of a plurality of voxels one of two or more predetermined labels; and
extracting a connecting boundary between each voxel based on the assigned predetermined labels of each voxel using an image processing technique to fit a plane.
22 . The computerized method of claim 21 , wherein deriving anatomical measurements based on the one or more identified anatomical landmarks comprises deriving a closed curve indicative of a boundary of the anatomical structure in the plane to calculate at least one of area-based, perimeter-based, minimum, and maximum diameters of the one or more identified anatomical landmarks.
23 . The computerized method of claim 1 , wherein deriving anatomical measurements based on the one or more identified anatomical landmarks comprises deriving anatomical measurements of the one or more identified anatomical landmarks from the virtual three-dimensional model of the anatomical structure.
24 . The computerized method of claim 1 , wherein the identified anatomical landmarks are overlaid on the anatomical structure in the virtual three-dimensional model of the anatomical structure.
25 . The computerized method of claim 1 , further comprising:
receiving user input feedback based on the displayed virtual three-dimensional model; and
adjusting the anatomical measurements based on the user input feedback.
26 . A system for automated anatomical analysis of an anatomical structure, the system comprising:
a memory having instructions that when executed by a processor operatively coupled to a computer, cause the computer to:
obtain a plurality of images of patient-specific cardiovascular anatomy, said plurality of images comprising a plurality of voxels;
analyze the plurality of images with a trained artificial intelligence module to identify one or more anatomical landmarks and to construct a virtual three-dimensional model of the anatomical structure, wherein to analyze the plurality of images with the trained artificial intelligence module comprises:
(i) generate a probability mask, the probability mask comprising a probability value for each voxel of the plurality of voxels, the probability value indicating a probability that said voxel corresponds to the anatomical structure;
(ii) convert the probability mask to a segmentation mask by comparing each probability value to a threshold value on a voxel-by-voxel basis to assign each voxel of the plurality of voxels a predetermined label corresponding to the anatomical structure if such probability value exceeds the threshold value; and
(iii) generate a segmentation mask comprising each voxel of the plurality of voxels assigned the predetermined label; and
(iv) construct the virtual three-dimensional model of the anatomical structure based on the segmentation mask;
derive anatomical measurements based on the one or more identified anatomical landmarks;
obtain specifications of a plurality of implantable devices associated with the anatomical structure;
display a sizing chart based on the specifications of the plurality of implantable devices and on the anatomical measurements;
receive input from a user comprising a selection of an implantable device from the plurality of implantable devices; and
display the virtual three-dimensional model of the anatomical structure alongside the anatomical measurements of the one or more identified anatomical landmarks and a virtual model of the selected implantable device.
27 . A non-transitory computer readable medium programmed with instructions that, when executed by a processor of a computer, cause the computer to:
obtain a plurality of images of patient-specific cardiovascular anatomy, said plurality of images comprising a plurality of voxels;
analyze the plurality of images with a trained artificial intelligence module to identify one or more anatomical landmarks and to construct a virtual three-dimensional model of the anatomical structure, wherein to analyze the plurality of images with the trained artificial intelligence module comprises:
generate a probability mask, the probability mask comprising a probability value for each voxel of the plurality of voxels, the probability value indicating a probability that said voxel corresponds to the anatomical structure;
convert the probability mask to a segmentation mask by comparing each probability value to a threshold value on a voxel-by-voxel basis to assign each voxel of the plurality of voxels a predetermined label corresponding to the anatomical structure if such probability value exceeds the threshold value; and
generate a segmentation mask comprising each voxel of the plurality of voxels assigned the predetermined label; and
construct the virtual three-dimensional model of the anatomical structure based on the segmentation mask;
derive anatomical measurements based on the one or more identified anatomical landmarks;
obtain specifications of a plurality of implantable devices associated with the anatomical structure;
display a sizing chart based on the specifications of the plurality of implantable devices and on the anatomical measurements;
receive input from a user comprising a selection of an implantable device from the plurality of implantable devices; and
display the virtual three-dimensional model of the anatomical structure alongside the anatomical measurements of the one or more identified anatomical landmarks and a virtual model of the selected implantable device.
28 . A computer program product comprising code portions with instructions that, when executed by a processor of a computer, cause the computer to:
obtain a plurality of images of patient-specific cardiovascular anatomy, said plurality of images comprising a plurality of voxels;
analyze the plurality of images with a trained artificial intelligence module to identify one or more anatomical landmarks and to construct a virtual three-dimensional model of the anatomical structure, wherein to analyze the plurality of images with the trained artificial intelligence module comprises:
generate a probability mask, the probability mask comprising a probability value for each voxel of the plurality of voxels, the probability value indicating a probability that said voxel corresponds to the anatomical structure;
convert the probability mask to a segmentation mask by comparing each probability value to a threshold value on a voxel-by-voxel basis to assign each voxel of the plurality of voxels a predetermined label corresponding to the anatomical structure if such probability value exceeds the threshold value; and
generate a segmentation mask comprising each voxel of the plurality of voxels assigned the predetermined label; and
construct the virtual three-dimensional model of the anatomical structure based on the segmentation mask;
derive anatomical measurements based on the one or more identified anatomical landmarks;
obtain specifications of a plurality of implantable devices associated with the anatomical structure;
display a sizing chart based on the specifications of the plurality of implantable devices and on the anatomical measurements;
receive input from a user comprising a selection of an implantable device from the plurality of implantable devices; and
display the virtual three-dimensional model of the anatomical structure alongside the anatomical measurements of the one or more identified anatomical landmarks and a virtual model of the selected implantable device.