Resumen
A system for determining the prognosis of a patient suffering from pulmonary embolism is provided. The system may include at least one computer system configure to receive patient specific data regarding his pulmonary embolism status. The at least one computer system may be further configured to create a model of the patient's heart, with at least information of the two ventricles, and to determine the ratio of sizes of the ventricles. The system will then report such ratio to the clinician or report a risk index of clinical outcome for such patient.
Reivindicaciones
1. CLAIMS 1. - Method for determining the prognosis of a patient suffering from pulmonary embolism, the method comprising: - receiving a plurality of axial images of the patient showing at least the heart, forming a three-dimensional image formed by voxels; - estimating in at least one three-dimensional image the location and shape of the contour of the ventricles of the heart of the patient; - estimating in at least one three-dimensional image the location of the inter-ventricular septum of the heart of the patient modeled as a plane; - estimating the diameter of the ventricles of the heart as the maximum width of the contour of each ventricle measured in lines perpendicular to the interventricular septum; - reporting the value of the division of the maximum measurement for a right ventricle to a maximum measurement for a left ventricle. determining the prognosis of the patient if the reported value of the division is higher than a predefined threshold. 2. - Method according to claim 1, wherein estimating in the image the location and shape of the contour of the ventricles further comprises: - providing an axial slice formed as a 2-dimensional model of the shape of the right ventricle and an axial slice as a 2-dimensional model of the shape of the left ventricle by training a machine learning based algorithm; - detecting in each axial slice the shape and location of the right ventricle and the shape and location of the left ventricle by said trained algorithm; - for each detected ventricle in each axial slice, providing a 2-dimensional bounding box and a score representing the fitness of the detection to each model. 3. - Method according to claim 2, wherein it further comprises: - selecting detected ventricles that are coherent with size and locations of 2- dimensional models previously recorded when modeling; - clustering detected ventricles according to the similarities on position of the detections, their sizes and their aspect ratio; - ranking clusters and selecting the cluster with the highest score for the left and the right ventricle; - for the left and for the right ventricles of the selected cluster, locate a seed point in the axial slide related to the centroid of said selected ventricle, preferably being the centroid. 4. - Method according to claim 3, wherein detected ventricles are coherent with size and locations of models if: - absolute value of the difference between the area of the detected ventricle and the model is less than a predetermined predefined value; and, - absolute value of the difference between the position of the mass centre of the image and the centroid of the detected ventricle is less than a predefined value. 5. - Method according to any previous claim, wherein the location of the inter-ventricular septum of the heart modeled as plane which is located by: - for each pair of seeds, the seed of the right ventricle and the seed of the left ventricle, the line connecting both seeds is computed; - for each line, calculating the location of the voxel wherein the three-dimensional image intensity is a maximum if the intensity of the septum is higher than the intensity of the ventricle, or minimum if the intensity of the septum is lower than the intensity of the ventricle; - defining a volume comprised by the voxels previously calculated; - computing the plane that represent the inter-ventricular septum form from the plurality of voxels of such volume, wherein said computation comprises: o for each voxel, computing the hessian matrix of the three-dimensional image at such voxel location being the hessian matrix <img class="EMIRef" id="284260395-imgf000028-0001" /> o wherein H<tj>, i,j = x, y, z represents the convolution of the image with the second derivative along the dimensions i,j of a gaussian kernel of a predetermined sigma; o for each voxel, computing the eigenvalues and eigenvectors of the associated hessian matrix; o for each voxel determining a score proportional to a plate-like structure traversing said voxel; o selecting a predetermined number of voxels within the volume with the highest score; o calculating the median direction of the eigenvectors of the selected voxels, being said median direction the direction of the estimated septum; o defining the plane of the septum as the plane containing the voxel whose eigenvector is closest in direction to the median direction and its normal being the median direction. 6. - Method according to claim 5 and any of previous claims, wherein each ventricle is segmented, by evolving it by a contour form from at least one seed point by means of a level-set algorithm under constraints over the three-dimensional image. 7. - Method according to claim 6, wherein the three-dimensional image constraints are defined by a learning pre-process of the level-set algorithm running it over a plurality of cases taken from an evaluation dataset. 8. - Method according to claims 2 and 5, and any of previous claim, wherein estimating the diameter of the ventricles further comprises: - for each slide, determining the septum line as the intersection between the plane of the septum and the slide; - defining the cross measurement of the ventricle as the distance value between the closest point of the contour and the farthest point of the contour along a line perpendicular to the septum line at certain x<t>of said septum line if both points exist; - defining a discrete function p<t>= f(x ), i = 1. . n for a predetermined number of sample points n wherein x<t>are points along the septum line having cross measurement being sorted from the apex to the atrium and p<t>said cross measurement value; - fitting the discrete function p<t>= f(xt), i = 1. . n by a polynomial of order n; - determining the estimated diameter or caliper as the value of the polynomial at the at least one root of its first derivative. 9. - Method according to claim 8, wherein the estimation of the right ventricle diameter further comprises: - defining a closed domain of the polynomial as the domain where the cross measurement exists; - calculating the first derivative p<n--1>'( ) from p<n>(x) and its roots (rpd); - calculating the second derivative p<n>_<2>"(X) from p<n>( ) and its roots (rpdd); - selecting the root of the polynomial for the estimation of the diameter of the right ventricle according to the following criteria: o If there are three or more roots in p<n--1>'( ) and p<n>_<2>"( ) on the first root, being the roots sorted from the left boundary to the right boundary of the closed domain, is lower than 0 and p<n>_<2>"(X)<on tne>second root is grater than 0 and the p<n>_2"( ) at the third root is lower than 0, take as root of the polynomial for the estimation of the diameter of the right ventricle the first <img class="EMIRef" id="284260395-imgf000030-0001" /> o else, if there are at least two roots in p<n>_i'(X) and the p<n>_<2>"(X)<on tne>f'<rst>root is lower than 0 and the p<n>_<2>" (x) on the second root is grater than 0, take as root of the polynomial for the estimation of the diameter of the right ventricle the position of the first root of p<n--1>'( ), o else, if there is at least one root in p<n>_<2>"(X), the value of p<n>_i'(X) at the first of these roots is greater than 0 and smaller than a predetermined threshold, take as root of the polynomial for the estimation of the diameter of the right ventricle the value of that root, o else, if there is at least one root in p<n>_i'(X) and p<n>_<2>"( ) in the first of the roots is lower than 0, take as root of the polynomial for the estimation of the diameter of the right ventricle the position of the first root of p<n>_<1>'( ). 10.- Method according to claim 8, wherein the estimation of the left ventricle diameter further comprises: - defining a closed domain of the polynomial as the domain where the cross measurement exists; - calculating the first derivative p<n>_i'(X) from p<n>(x) and its roots (rpd); - calculating the second derivative p<n>_<2>"( ) from p<n>(x) and its roots (rpdd); - If the value of p<n>_2"( ) at the first root of p<n>_i'(X) is lower than 0, take as root of the polynomial for the estimation of the diameter of the left ventricle said first root of 11.- Method according to claims 9 or 10, wherein after calculating the roots of the first derivative p<n>_i'(X) and the roots of the second derivative p<n>_2"( ), those roots closer than a predetermined value to the extremes of the domain are removed. 12.- Method according to any of claims 8 to 11 and any of previous claims, wherein the maximum distance value between the closest point of the contour and the farthest point of the contour along a line perpendicular to the septum is estimated as the maximum value of the maximum of the polynomials computed at each axial image.
