WO1997048978A1

WO1997048978A1 - Quantitative coronary artery analysis system

Info

Publication number: WO1997048978A1
Application number: PCT/US1997/010675
Authority: WO
Inventors: Michelle T. Lefree; Joseph Sitomer
Original assignee: Quinton Instrument Company
Priority date: 1996-06-21
Filing date: 1997-06-19
Publication date: 1997-12-24
Also published as: AU3573297A; CA2230027A1

Abstract

A quantitative coronary artery analysis system (30) is described which includes a determination (40) of the set of most likely data points representing the edge of an artery to enable an analysis program to calculate the dimensional information (46) of interest for a patient from an x-ray image of the selected artery wherein the program includes the steps of performing an axi-linear transform (38) of the image data with subsequent filtering, rejection and correction of selected data points (42) such that the user is provided with the calculated dimensional data (46) for an artery, other blood vessel, or organ of a patient.

Description

QUANTITATIVE CORONARY ARTERY ANALYSIS SYSTEM

Field of the Invention

The present invention relates generally to an improved system for the calculation of the diameter of blood vessels or other organs of interest in a patient using an imaging system and more particularly to an improved system for performing coronary artery analysis and even more particularly to an improved system which may use a novel arterial phantom having known internal diameters in combination with an angiographic catheter and an improved image quality phantom to increase the accuracy of the quantitative coronary artery analysis programs.

Background of the Invention

During angiographic or other diagnostic procedures which use X-Rays, a system similar to that shown generally in Figure 1 is used. The system generally includes an X- Ray tube 10 which emits X-Rays from a general point source. The X-Ray tube is positioned under a table 12 on which the patient or other object of interest is positioned. In order to provide for X-Ray photography, cinegraphic recording and/or viewing of the subject, an image intensifier 14 is positioned above the subject.

The bottom surface of the image intensifier 14 includes a grid 16 having a plurality of narrowly spaced strips thereon in order to attenuate scattered X-Rays so that only the X-Rays which pass through the subject directly from the X-Ray tube. The top of the image intensifier includes a TV or similar camera 18 to dynamically view the subject of the study. For example, the heart or other organ of a patient may be viewed through the camera to allow the physician the ensure the proper positioning and observe the operation heart or other organ. Additionally, the top portion of the image intensifier 14 also typically includes a recording medium 20 such as a film camera or digital recording medium to record the study for later review and analysis. The image intensifier incorporates a phosphorous screen 22 and a series of focusing coils 24 which tend to cause the X-Ray beams to be directed toward a phosphorous output lens 26. The beam from the output lens is split by a beam splitter 28 to provide output to the TV camera 18 and recording medium 20.

Current imaging systems use various enhancement techniques to increase the visibility of various structures in an image. This "front end" enhancement increases the difficulty in attempting to analyze the image because the enhancement decreases signal-to-noise ratio in the image and distorts the image by increasing the arterial noise by creating a halo or ringing along the arterial structure. Current systems which are used to analyze coronary arteries during diagnostic angiographic procedures include automatic analysis programs to calculate the dimensions of the arteries of a patient. While these programs have been used for many years, the programs suffer from certain inaccuracies which result from distortions that occur during the image acquisition process as well as limitations in the sharpness of the pixels of the acquired image. For example, various studies have shown that the accuracy of the analysis programs decreases as the size of the artery of interest decreases and it is the arteries having the reduced sizes that are of particular interest to the cardiologist. Despite these limitations, there is also a need to improve the underlying analysis program while understanding the limitations of the data received by the analysis program.

In one such currently available edge detection analysis program, the quantitation software was designed to be used for analyzing cinefilm images and has been adapted for use during the course of a diagnostic catheterization or a PTCA procedure. As such, the analysis program is designed to require minimal operator interaction and to provide results rapidly enough to be of use during the procedure. This feature is particularly desirable in PTCA because live time feedback concerning the status of an arterial lesion is important to determine the effectiveness of the therapy. The operator of this analysis program then identifies the lesion of interest and identifies several estimated center points.

The currently available analysis program then calculates the arterial centerline and begins to determine the arterial borders (edge detection) . First the analysis program identifies potential or preliminary edge points using Laplacian filtering to calculate the second derivative. The second step includes a process to discriminate between the actual edge of the desired object in the image and the potential edges due to the presence of adjacent structures or image noise. In this currently available analysis system, rather than applying a global edge enhancement operator to the entire image, a series of linear intensity profiles which are perpendicular to the arterial centerline are extracted and processed in two passes. This action increases the efficiency of the edge detection process by reducing a two dimensional signal processing problem to one of a single dimension. This application of a-priori knowledge as to the location of the arterial segment of interest, as well as the discrimination against "edges" due to the presence of adjacent or superimposed radiographic structures or image noise are also implemented using the current technique.

The actual edge detection of the currently available analysis system begins by identically processing each of the previously mentioned linear perpendicular profiles on a first pass over the arterial centerline. A one dimensional Laplacian convolution filter is applied to approximate the second derivative of the given profile. The resulting second derivative is examined to locate the position of its maxima and zero crossings. The actual examination process is simplified knowing the position of the geometric center of the artery along the profile. The 97/48978 PC17US97/10675

- 5 -

search is begun at the geometric center of the artery and continues outward in either direction until the positions of the maxima and zero crossings are found. As used herein and shown generally in Figure 2, the term "maxima" is used to refer to the arterial base points and the "zero" crossings are the arterial inflection points.

Both the arterial base points and inflection points are used by the currently available analysis program because radiographic experiments show that the inflection points yield edges which are too narrow with respect to the artery and the base points yield edges which are too wide. In order to modulate the position of the detected edge to lie somewhere between these points, the commercially available analysis program utilizes information based on the image density at these locations. Therefore, the commercially available analysis program uses a variable threshold between the densities at the base and inflection points on either side of the artery which includes an optimum value for this threshold of seventy-five percent (75%) of the density difference at the base and inflection points.

Each perpendicular profile is processed in this manner during the first pass over the arterial segment. The majority of the information used in the first pass is based on knowledge of the slope of the arterial perpendicular profile. The second pass uses a combination of threshold information and a-priori knowledge concerning the spatial continuity of the coronary arteries. As mentioned previously, a major part of the general edge detection of this analysis program is based on the desire to discriminate against spurious edges due to image noise or the presence of adjacent or superimposed radiographic structures.

In edge detection, it is often the case that edges determined using information regarding the magnitude of the gradient at the arterial border alone may form noisy edge contours. This is particularly true near highly significant stenoses, where the arterial gradient is often smaller in magnitude than the gradient due to image noise in adjacent regions. Additionally, side branches or parallel structures may cause jump discontinuities in the calculated arterial border. In order to overcome these problems, the commercially available analysis program employs a spatial continuity criterion after the first pass to discard points which fall an empirically determined epsilon distance from neighboring edge points.

In the second pass of this analysis program, each perpendicular profile is searched to locate the position of the densities which were determined for a given profile on the first pass. The points which are discarded by the spatial continuity criterion are treated differently than the valid edge points which have passed the criterion. In those instances, the second pass involves a similar outward search along the given profile, except that the density searched for is calculated as an average of densities at valid neighboring valid points. Therefore, this analysis program uses both gradient and threshold information, as well as a-priori knowledge concerning the spatial position of the artery itself and information regarding the spatial "smoothness" of arterial structures. The edge contours provide projected geometric diameter information (in pixel units) along the arterial segment defined by the operator. In order to calibrate the dimension of the pixels at that distance from the image intensifier, the magnification factor must be known. The commercially available analysis system uses the procedure catheter to make this determination. This step assumes that the artery segment and procedure catheter are at the same distance and uses the known catheter diameter, i.e. french size, relative to its measured pixel diameter for this determination. In clinical practice, the operator selects a small segment of the procedure catheter and the same edge detection routine as described above is used to calculate the magnification correction factor.

Based on the foregoing, there remains a need for an improved analysis program which recognizes the weaknesses of current imaging procedures and overcomes much of the variability in the data that is presented to the analysis program.

Furthermore, there remains a need for a reliable and consistent analysis program which allows the image data from an imaging system to be analyzed to identify multiple possible edges and then determine which of the possible edges are most likely to be the actual artery edge. Summary of the Invention

An advantage of the present invention is that it reduces the side effects of current front end edge enhancement techniques whereby quantum noise is enhanced and arterial noise is enhanced to provide a more accurate calculation of the dimensions of the blood vessel than is available with currently available analysis programs.

Another advantage of the present invention is that it is based on the recognition that radiographic images are random fields by nature which contain anatomic projections and that the information contained in these projections is fundamentally stochastic in nature and that the mathematics of random variables are an appropriate tool to deal with this type of information content. Yet another advantage of the present invention is that the analysis program performs an axi-linear transform on the artery section of interest to create fixed, reproducible pixel dimensions at high resolution of about 20 pixels per millimeter and allows for consistent and intelligent axial and radial filtering of the data to compensate for noise in the arterial image.

Yet another advantage of the present invention is that the analysis program provides for a first pass edge point estimation that is inflection point based and uses probability density functions to determine the maximum likelihood of the position at actual arterial edge. The present invention utilizes an analysis program wherein once the user identifies the centerline of the segment of the artery or image of interest, the analysis program refines the identified centerline. The user then has the ability to edit the calculated centerline. Next, the analysis program performs an axi-linear transform of the image to transform the image data from a data set which includes curved arterial portions therein to a generally linear depiction of the artery of interest. The axi-linear transform step takes the curvilinear axis of the artery and the perpendicular profiles as shown in Figure 3 and straightens the axis so that the perpendicular profiles are arranged equidistantly from each other. Additionally, this step interpolates the image data up to about 20 pixels/mm. As a result of this transform, the image data may be filtered along the x and y axis to correlate information from adjacent perpendicular profiles; and because the bandwidth of the artery is relatively fixed, interpolation of the image data to about 20 pixels/mm from about 4 pixels/mm filtering may be used to remove image data having a bandwidth less than about 10 pixels without affecting the ability of the analysis program to analyze artery segments having diameters of 0.5 mm. The analysis program then performs a first pass analysis of the data to determine the points with the most likelihood of lying on the actual arterial border. In this first pass, the features of the perpendicular profiles from the axi-linear transfer are then parameterized based on: 1) the distance from a potential boundary point to a local edge trend as defined by the directional IIR filtered distance from the edge points identified in the adjacent perpendicular profile; and 2) on the amplitude of the first and second derivatives of the perpendicular profile. The parameters are then sampled at the second derivative extrema to create probability density functions. In this step, data is extracted which corresponds to the features of interest to reduce the data to a few decimated data points. The maximum likelihood operators are then applied to the probability density functions to isolate the most likely inflection point. The results of the first pass are then evaluated in a second pass boundary refinement step to determine whether or not certain points are outliers that do not fit with nearby points and to perform a modest regression correction. As part of the outlier rejection step, the differential radius and intensity of the data is determined as a function of the axial position of the data points. Global first order statistics are used to reject outliers because radial outliers are believed to indicate noise points while intensity outliers are believed to indicate extraneous vasculature. Finally, the analysis program also performs the second pass over the data using the regression data from the particular imaging system to modulate the diameter-wise corrections about the first pass, inflection point based edges. This allows the selected edge point to be moved up or down the artery profile depending on the desired amount of correction for certain inflection points.

One of the fundamental premises of the present analysis program is based on the knowledge that the formation of angiographic images is a stochastic process. Therefore, the present invention uses the mathematics of random variables as part of the edge detection determination to address both the structured and unstructured noise present in the image. Several probability distribution functions are defined which parameterize the placement of arterial edge points along image density vectors normal the arterial medial axis. The parametric edge descriptors are the magnitudes of the first and second directional derivatives of the image density vectors and the edge trend which is the IIR filtered radial distance from the position of the edges of adjacent perpendicular profiles to the extrema of the first and second derivatives of the image density vectors. Maximum likelihood techniques are applied to combine these parametric edge descriptors as probability distribution functions in determining the initial placement of chord- wise arterial edge points. The two sets of these maximum likelihood determined edge points comprise preliminary edges along either side of the artery. The remainder of the analysis program is directed at compensating for the effect of the statistical quality of the particular angiographic image on these preliminary edges and identifying and avoiding arterial side branches or over/underlying vasculature. Global first order statistics of the parametric edge descriptors and image densities evaluated along the preliminary edges are calculated to distinguish between valid arterial edge points, deviations caused by correlated and/or uncorrelated image noise. In this context, uncorrelated image noise generally refers to random noise while correlated noise may be caused by overlying, underlying or adjacent vasculature or by the ringing or halo effect of the enhancement feature of the imaging system or the presence of extraneous vasculature. The values of the edge descriptors at individual preliminary edge points are compared with the global statistical measures to discriminate against false or outlying edge points. The concordance or discordance of statistical deviations in the parametric edge descriptors versus deviations in image densities along the preliminary edges indicate the presence or absence of the effects of image noise and/or extraneous vasculature.

An advantage of the present invention includes the significantly improved robustness relative to the deleterious effects of front end edge sharpening techniques on image quality in modern angiographic imaging systems.

Another advantage of the present invention is the enhanced ability to discriminate against overlying, underlying or adjacent vasculature.

Yet another advantage of the present invention is that the improved robustness of the present invention against image noise minimizes the need in current analysis programs for aggressive low pass filtration thereby resulting in an improved diameter response curve for the detection of diameters of less than 1 mm.

Yet another advantage of the present invention is the ability to adjust the zoom capabilities of the analysis program so that the entire portion of the artery of interest may be analyzed based on individual variations within the selected portions of the artery of interest.

Brief Description of the Drawings

Figure 1 is a schematic drawing of a typical imaging system;

Figure 2 is an illustration of an artery profile curve;

Figure 3 is an illustration of the artery profile and analysis lines of the prior art analysis system;

Figure 4 is an illustration of the axi-linear transform step of the present invention; and

Figure 5 is a general flow chart of the components of the present invention.

Detailed Description of the Present Invention

As described above, the present invention applies the mathematics of random variables to the image data received from an imaging system. As shown in Figure 2, the typical arterial profile is a generally bell shaped curve when it is aligned perpendicular to the centerline of the artery. Bl and Br are the left and right base points and II and Ir are the left and right inflection points of the arterial profile. Figure 1 diagrammatically illustrates the connection of the analysis program 30 to the imaging system. The analysis program 30 may be directly connected to the data from the imaging system via a computer as shown or may be performed as part of a later analysis of the angiography procedure to document the success of the prior procedure. The present invention includes the general steps of manually selecting the portion of the artery of interest and drawing a centerline 32 on the selected portion of the artery using a computer mouse or similar line drawing approach to define individual points approximately at the center of the artery. This step functions to identify the location of the arterial segment to be analyzed and reduces the amount of data which must be analyzed by the analysis program. Next, the analysis program 10 performs an initial analysis of the points and connects the selected points to draw an approximate centerline 34 based on the user selected points. The user may then move or edit portions of the line 36 to more accurately identify the approximate centerline of the artery. The next steps in the analysis of the selected portions of the arteries of interest are performed internally by the analysis program. The first step is performed once the approximate centerlines are identified and verified. In this step, the analysis program mathematically performs an axi-linear transform 38 of the selected portion of the artery. The axi-linear transform 38 takes into account the image intensity of the selected portion of the artery and provides a relatively high resolution of about 20 pixels per millimeter. The analysis system then operates on the axi-linearly transformed 38 image by performing a maximum likelihood analysis 40 on the image data retained therein. The first step of the analysis program is to filter along and normal to the axis of the artery segment to limit the frequency content of the transformed image independently in those directions. Because of this, any frequency content which represents a structure smaller than the smallest coronary artery size, such as about 0.5 mm, may be removed as noise. The image density vectors which are oriented normally to the arterial medial axis of the selected portion of the artery are then parameterized to identify the most likely edge points from the various points of data from the image data for the selected profile. The parameters include the magnitudes of the first and second directional derivatives of the profiles and the IIR filtered radial distances from the edge trend to the extrema of the first and second image density vectors. The possible identified edges are then evaluated according to various criteria to determine the data point which is most likely the actual edge point. The criteria used for this evaluation include: 1) the closeness of the data point to prior determined edge points; 2) the steepness of the slope at the inflection point of the arterial profile as illustrated by the first derivative of the perpendicular profile; and 3) the flatness of the valley between the peaks as illustrated by the second derivative of the perpendicular profile.

Next, the analysis program performs an evaluation of the first pass analysis of the data to remove the outlying data points 40. In this step, the preliminary edge data, which consists of a set of inflection based edge points, is evaluated to determine if outlying data points are present. In the event outliers are identified, they are replaced in the following manner. Analysis is restricted to a limited neighborhood along the profile in question. The neighborhood is determined around the location of adjacent valid edge points which have passed the outlier rejection test. Replacement points are determined by searching the neighborhood for a pixel intensity similar to intensities of neighboring valid edge points.

Finally, the selected edge points are subjected to the regression curve 44 to correct the data based on known biases or distortions of the specific imaging system. The regression data may be obtained from a variety of sources including an approach wherein the tip of a procedure catheter which is the same as or similar to the procedure catheter to be used in the procedure is imaged along side a phantom which is filled with an imaging material that is similar to the imaging material to be used in the procedure. The phantom preferably includes a telescopic shaped, dye filled internal area having various sections of known external diameter. From this initial imaging step, the user is able to compare the imaged size of the procedure catheter as well as obtain calibration data for the regression curve.

Thereafter, the selected edge points are presented to the user for their review and potential editing 46. When the user approves the selected points or finishes the editing of the selected points, the analysis program calculates the relevant diameter and cross sectional diameter of the selected artery segment. As part of the usefulness of this information, the user may identify various points along the selected artery and the analysis program will identify the relevant dimensions of the selected points. This type of information allows the user to determine the severity of blockage at the selected location or the amount of improvement following a therapeutic procedure.

Claims

QUANTITATIVE CORONARY ARTERY ANALYSIS SYSTEMClaimsWhat is claimed is:

1. A system for the analysis of the imaging data from an imaging system to determine the dimensions of a blood vessel or other organ of interest, the system including: a computer including a means for receipt of imaging data from an imaging system, a processor for processing the imaging data to perform an axi-linear transfer of the imaging data and filtering the data to remove noise and distortion from the imaging data; and a display means to display the transferred and filtered imaging data for use by the user.

2. The system of claim 1 wherein said processor performs an analysis of the imaging data to determine the maximum likelihood that a set of data points represents the boundary of the blood vessel or other organ of interest.

3. The system of claim 2 wherein the imaging data is data resulting from the X-Ray imaging of an artery of a patient and said set of data points is an arterial boundary.

4. The system of claim 2 wherein said processor performs statistical outlier rejection of the selected data points and adjusts the selected data points based on a predetermined arterial diameter-wise data regression table.

5. A method of determining the dimensions of a blood vessel or other organ of interest using an analysis program which analyzes image data from an imaging system, including the steps of; receiving image data from an imaging system; performing an axi-linear transform of the imaging data; determining the maximum likelihood that particular points from the imaging data are the points of interest; rejecting statistical outliers from a set of data points of interest based on predetermined criteria; performing a regression correction on the remaining set of data points; and displaying the set of data points for use by the user of the analysis program.

6. The method of claim 5 wherein the image data is from an X-Ray system and the image data is from an X-Ray image of an artery wherein the method includes the step of determining the most likely location of the edge of the artery.

7. The method of claim 6 including the step of determining the most likely location of the edge of the artery from a set of data points which are selected based on criteria that indicates that a particular data point is a possible edge data point.