EP0630503A1 - Identification of anatomical features from data - Google Patents

Abstract

An image obtained in nuclear medicine is scanned and pixels of maximum and minimum intensity are identified and correlated with each other using constraints which are empirically determined to relate to the characteristic to be studied (such as the heart ). The information thus obtained is used to define a region to be studied in which an anatomical characteristic to be studied can be found, and to position a detector with a scintillation camera allowing a nuclear medicine study to be carried out in optimal positions. In order to correctly position a patient, anatomical landmarks relating to a target organ are automatically identified by a computer. The markers are superimposed on a persistent image of the target organ. This makes it easier for a technician to identify the target organ, as well as to repeat studies such as myocardial perfusion studies for which two studies must be performed on a single patient at two different times. An anatomical landmark associated with the target organ is identified by computer in each of the images, and the depth of the target organ is geometrically determined using the differences in size between the images of the identified landmark, as well as differences in height. .

Description

-EDENπFICATIO OF ANATOMICAL FEATURES FROM DATA

Technical Field

The invention relates to nuclear medicine and, more particularly, to images produced by nuclear medicine studies of patient organs. In its most immediate sense, the invention relates to nuclear medicine studies of the heart, proper positioning of a patient in studies of patient organs, and determinations of the depth of an organ, such as the heart, within the body. Background Art

In contrast with nuclear medicine studies, modalities such as computed tomography and magnetic resonance imaging produce clearly defined images. However, this is not the case in nuclear medicine, e.g. cardiac bloodpool studies - - which are concerned more with functional determinations such as ejection fractions rather than the location of malignant regions as in other studies. Here, since cardiac tissue is engorged with blood and the tissue adjoins blood pools within the heart, there are no sharp edges (as exist with images produced by other modalities) and even under the best of circumstances it is difficult to locate exactly that portion of the image which relates to, e.g. , the left ventricle. Thus, it is necessary to identify such organ features as the interventricular septum, the left ventricle, and the right ventricle.

Attempts have been made to use edge detection methods to identify the location of anatomical features, but such attempts have been unsuccessful.- This is because the data used to construct such images is contaminated by scatter and attenuation. Additionally, nuclear medicine images become even more ambiguous because of low resolution, low signal-to-noise ratio, and the presence of radioactivity from adjacent organs and background tissue. Consequently, conventional detection methods are not helpful in identifying anatomical structures of interest; these methods may locate an edge which is without diagnostic significance and may overlook an edge which is highly important.

Additionally, it frequently happens that the camera is mispositioned with respect to the patient's target organ (i.e., the organ of interest). (This may happen where the patient has, e.g., an abnormally oriented heart, or where the patient shifts position after being positioned properly.) Even an experienced technician needs substantial time to collect enough data to assess the positioning of the patient and to correct any mispositioning, and an inexperienced technician may need to repeat these steps one or more times before the patient is positioned properly. Worse, if the mispositioning is not detected and a study has been conducted with the patient in a suboptimal position, the diagnostician is forced to choose between using the resulting suboptimal study (which is of diminished diagnostic utility) or repeating it (and thereby dosing the patient once again and decreasing throughput through the camera) . Another concern relates to positioning of the apparatus with respect to the body. In some nuclear medicine studies, such as bone scans, it is easy to properly position a patient with respect to the detector of a scintillation camera. However, in other studies, particularly studies of organs such as the heart, such proper positioning is difficult because many relevant factors (e.g. orientation of the heart within the patient's body, size of the various portions of the heart) are not known before the study is commenced. In these latter instances, a nuclear medicine study is usually commenced by positioning the detector of the scintillation camera in approximately the correct orientation, verifying the correctness of the position originally selected, and adjusting the position as necessary. The verification step is usually carried out by viewing a persistence image of the Region Of Interest (ROI) on the camera display. There are two major disadvantages to this approach.

First, when using a focussing collimator, it can be difficult for a technician to identify the anatomical feature of interest (for example, the patient's left ventricle) because its appearance can be quite distorted by the focussing scheme. Second, even if the anatomical feature of interest is identified, the positioning is not highly repeatable. Thus, for example, in cardiac myocardial perfusion tests, it is advantageous to conduct a first study with the patient at rest and a second study after stressing the patient. The patient should be in the same position in both studies. If the location of, e.g., the heart within the ROI is only known in a general way, the patient's position during the second study will not precisely replicate the patient's position during the first study.

It is similarly advantageous for a diagnostician to know the depth of an organ which is the subject of a nuclear medicine study. For example, in nuclear medicine studies of the renal function, physicians evaluate this function by conducting a planar image study and comparing uptake of radioisotopes with results that are tabulated by body size and depth of the kidney within the patient.

Such depth information cannot easily be determined from planar nuclear medicine image data. If such depth information is to be obtained using SPECT (single-photon emission computed tomography) data, it is necessary to conduct a second study, and SPECT studies are time- consuming and expensive.

It would therefore be advantageous to provide method and apparatus which would be capable of automatically identifying anatomical- features in nuclear medicine images for the target organ in nuclear medicine studies even where the target organ is not sharply defined.

It would also be advantageous to provide method and apparatus which could automatically identify the camera positioning between the patient's target organ and the camera detector far in advance of the end of the study, thereby permitting manual or even automatic repositioning of the camera head to an optimal position.

It would be advantageous to help a technician identify an anatomical feature of interest, even when the feature appears in a distorted manner because a focussing collimator is being used to conduct the study.

It would also be advantageous to provide method and apparatus which would make patient positioning more repeatable. It would be advantageous to be able to obtain organ depth information from planar nuclear medicine image data, without the need to conduct an additional SPECT study.

One object of the invention is to provide method and apparatus which can identify anatomical features in nuclear medicine images of target organs.

Another object of the invention is to provide method and apparatus which can automatically identify anatomical landmarks in nuclear medicine images, even when the images contain insufficient data to be diagnostically useful. A further object of the invention is to provide method and apparatus which will permit automatic positioning of the scintillation camera detector.

Another object of the invention is to provide method and apparatus which will help a technician identify an anatomical feature of interest, even when the feature is displayed in a distorted form.

Another object of the invention is to provide method and apparatus which makes patient positioning more repeatable. Yet another object of the invention is to permit organ depth information to- be obtained from planar nuclear medicine image data. Still a further object is, in general, to improve on known methods and apparatus in nuclear medicine. Disclosure of the Invention

The invention proceeds from the realization that conventional edge detection methods are unsuitable for use in nuclear medicine images. This is because such methods do not incorporate any anatomy-specific constraints and therefore cannot distinguish between patterns having diagnostic significance and patterns which lack diagnostic significance. Additionally, because nuclear medicine images of tissue are not sharply focussed even under the best of circumstances, it is difficult for edge detection methods to produce meaningful edge curves.

However, it is possible to take advantage of a basic characteristic of nuclear medicine in order to locate anatomical landmarks. In a nuclear medicine study, the uptake of a radioactive isotope determines the intensity of the nuclear medicine image and thus there are no shadows, reflections, or artificial highlights. Therefore, segments of maximum and minimum intensity will in certain instances be intrinsic to the patient's anatomy and will be unaffected by changes in imaging conditions. As a result, it is possible to utilize the relative locations of such maxima and minima in order to locate anatomical landmarks. In accordance with the invention, a nuclear medicine image is scanned line by line. For the intensity profile of each scan line, local curvatures are computed to identify local intensity maxima and local intensity minima. After the whole image has been so processed, line segments are constructed from all identified local intensity maxima, and other line segments are constructed from all identified local intensity minima. Then, these line segments are evaluated to see whether they satisfy constraints which are specific to the anatomical region of interest. If so, the line segments are treated as identifying particular anatomical landmarks, ,and on the basis of these landmarks, boundaries for the target organ or the portion of interest of the target organ can be easily located.

For example, if a cardiac image shows the left and right ventricles separated by the interventricular septum, there will be line segments constructed from pixels of maximum intensity near the long axis of each ventricle and there will be a line segment constructed from pixels of minimum intensity running near the interventricular septum and between the two line segments of maximum intensity. Furthermore, this "maximum-minimum-maximum" pattern will have a certain spatial relationship dictated by the structure of the heart. Therefore, "maximum-minimum- maximum" patterns which are heart-related can be distinguished from similar patterns which are not heart- related. Consequently, a computer can be programmed to scan a cardiac gated blood pool image and to draw the above-mentioned line segments (anatomical landmarks) .

It is possible, in accordance with the invention, to more easily deduce the boundaries of the ventricles of the heart from the above-mentioned line segments because there are certain geometrical relationships intrinsic to the heart and the boundaries can be expected to lie within regions defined by these anatomical landmarks. It is therefore possible to limit the identifications of anatomical features of interest to comparatively small regions which exclude anatomical structures that are not of interest.

Furthermore, because the referenced "maximum-minimum- maximum" pattern is dictated by the structure of the heart, that pattern exists at all stages of image acquisition even if there is insufficient data to form an image which would be diagnostically useful. There is a one-to-one relationship between the detected landmark pattern and the camera position. Thus, even after a relatively short time, the position of the camera detector with respect to the patient's target organ can be inferred from the relative positioning of the above-mentioned line segments. From this, it can be determined whether the camera has been properly positioned and how the camera should be repositioned; this permits the camera to be repositioned so that a study is not conducted at a suboptimal angle.

Additionally, a camera with a motor-driven gantry can 5 be adapted to automatically position itself to an optimal position. With such a gantry, the camera detector can be stepped around the patient at small angular increments (each step position is known as a camera stop) , with one frame of information being acquired at each position. In

30 further accordance with the invention, after relevant frames of information have been scanned, the inferred camera position is registered.

In accordance with another aspect of the present invention, the camera is set up at an initial position and

15 an image (advantageously a persistence image) is acquired. After a predetermined time, the data in the image is computer-analyzed in real time in accordance with the invention .described above and a set of position-defining anatomical landmarks of interest (e.g. the left ventricular

20 cardiac muscles of the heart) are thereby identified. (At least two such landmarks are necessary, in order to infer camera position.) A plot of such landmarks is then generated by the computer and the plot is then superimposed upon the part of the persistence image to which the plot 25 relates.

This permits the technician to immediately identify the significant portion of the image and to center (or otherwise locate) that portion on the display. Alternatively, the positioning of the significant portion

30 of the image can be detected by the computer and the technician advised of advantageous adjustments which might be made to the position of the patient and/or the camera

-ft detector.

In accordance with yet another aspect of the present 35 invention, a planar image of the organ of interest is acquired, using a focussing collimator, at two known and different heights. The result of this is two images which differ only in the degree of magnification. The method and apparatus for identifying anatomical features is then used to automatically determine, in each of the images, an anatomical landmark related to the organ of interest. Since the focal length of the collimator and the heights of the collimator are known, the difference between the size of the landmark in the two images permits the depth of the organ to be determined using simple geometrical relationships. Brief Description of the Drawings

The invention will be better understood with reference to the following illustrative and non-limiting drawings, in which:

Fig. 1 is a schematic diagram of a scintillation camera system in accordance with the preferred embodiment of the invention;

Fig. 2 is a planar image of a patient's heart taken at an angle of 45 degrees Left Anterior Oblique (LAO) ;

Fig. 3A is a persistence image of a patient's heart, formed using a parallel hole collimator;

Fig. 3B is a plot of the detected anatomical landmarks in the patient's heart, superimposed upon the image shown in Fig. 3A;

Fig. 4A is a persistence image of a patient's heart, formed using a focussing collimator;

Fig. 4B is a plot of the detected anatomical landmarks in the patient's heart, superimposed upon the image shown in Fig. 4A;

Fig. 5 is a flow chart in accordance with the preferred embodiment of the invention;

Fig. 6 shows lines which generally correspond to a patient's ventricles and interventricular septum, derived using the preferred embodiment of the invention;

Fig. 7 show how a region of interest for locating the left ventricle is determined in accordance with the preferred embodiment of the invention;

Figs. 8 and 9 show, respectively, a low-statistics

S image of a patient's heart taken at approximately 30 degrees LAO and the anatomical landmarks detected in the image in accordance with the preferred embodiment of the invention; Figs. 10 and 11 show, respectively, a low-statistics image of a patient's heart taken at approximately 50 degrees LAO and the anatomical landmarks detected in the image in accordance with the preferred embodiment of the invention; Figs. 12 and 13 show, respectively, a low-statistics image of a patient's heart taken at approximately 40 degrees LAO and the anatomical landmarks detected in the image in accordance with the preferred embodiment of the invention; Fig. 14 shows how the camera stop position affects the detected length of the interventricular septum as determined in accordance with the preferred embodiment of the invention;

Fig. 15 shows how the camera stop position affects the detected average intensity of pixels along the interventricular septum as determined in accordance with the preferred embodiment of the invention;

Fig. 16 shows how the camera stop position affects the detected end point position of the interventricular septum as determined in accordance with the preferred embodiment of the invention;

Fig. 17 shows the correlations between the relationships displayed in Figs. 14 - 16;

Fig. 18 is a flow chart in accordance with the preferred embodiment of the invention in which a scintillation camera detector is automatically positioned in a clinical environment;

Fig. 19 is a three dimensional graph of Fig. 2 in which image intensity is shown on the Z axis; Fig. 20 shows a horizontal scan line across the Fig.

2 image and the resulting localizations of local intensity maxima and minima based upon a local minimum and maximum

SUBSTIT curvature pattern;

Fig. 21 is a flow chart of a method in accordance with the preferred embodiment of the invention;

Figs. 22A and 22B show the geometrical relationships upon which the preferred embodiment of the invention relies; and

Fig. 23 is a flow chart in accordance with the preferred embodiment of the invention. Best Mode for Carrying Out the Invention Although the first aspect of the invention will be described below in connection with a cardiac bloodpool study, it should be understood that this is merely an exemplary application. The invention can be adapted to other applications by persons skilled in the art. In a conventional gated bloodpool study, a radioisotope is administered to the patient and the patient's heart is imaged at an appropriate position (typically, 45 degrees LAO). For determination of, e.g., the patient's ejection fraction, it is conventional to identify the boundaries of the patient's left ventricle at various points in the patient's cardiac cycle. (The details of gated bloodpool studies and computation of ejection fractions are not set forth here; they are known to persons skilled in the art.) Such an identification is difficult to make because the boundary of the left ventricle is not sharply defined in a nuclear medicine image. It is therefore necessary to use subjective judgment in determining where this boundary is located. Consequently, two different diagnosticians might have different assessments of a patient's ejection fraction even though both assessments are based on the same data. Indeed, since a single diagnostician may place the left ventricle at different locations at different times, that diagnostician may not even produce consistent assessments from the same data. It would obviously be advantageous to reduce-, as much as possible, the subjective element in such determinations. Additionally, because a gated bloodpool study consists of multiple frames of images (typically, 16 to 32 frames) , it is a tedious process to manually draw the boundary for each one of the frames of a study. It would obviously be advantageous to automate this process.

The following detailed description of preferred embodiments of the invention is based upon experiments conducted on 75 clinical examinations including more than 1000 frames of gated bloodpool images, which were acquired by different technicians in different hospitals on different types of heart conditions. The description concentrates upon identification of the boundary of the left ventricle because most quantitative assessments, such as ejection fraction, are referenced to it. However, as is described in more detail below, there is no reason why the invention could not be used to identify the boundary of the right ventricle.

Initially, and as is shown in Fig. 1 , a gated bloodpool study is conducted upon a patient 2. In such a study, a radioisotope is administered to the patient 2 and the heart 4 of the patient 2 is imaged using a scintillation camera generally indicated by reference numeral 6. A number of frames of planar images collected by the camera 6 is routed to a computer 8, and the planar image itself may be displayed upon a display 10.

In the normal case (see Fig. 2) the planar image is a two dimensional picture. However, it is also possible to treat the image as a three-dimensional graph, wherein the X and Y axes represent the Cartesian coordinates of a pixel in the image and the Z axis represents the intensity of the image at that pixel. (It will be understood that the intensity of the image at a particular pixel is a function of the number of detected scintillation events which relate to the location of that pixel in the patient 2.) Fig. 19 is such a version of the Fig. 2 image.

In accordance with the preferred embodiment of the invention - see Fig. 20 - the Fig. 2 image is scanned with a straight line which is parallel to the X axis. At any X position of this line, there will be a particular intensity value which can be treated as a Z value in the Fig. 19 graph, and there will be a plurality of local maxima and minima of Z. The local curvatures of the profile of this scan line (the profile of the scan line is the curve formed by all Z values as a function of X position) can be so computed that local minimum curvatures (negative peaks) precisely correspond to the local maximum points of the profile, the local maximum curvatures (positive peaks) precisely correspond to the local minimum intensity points of the profile, and so that the remaining points of the profile have zero curvature values. Thus, when such a computational scheme is utilized, the original nuclear medicine image is mapped to a local curvature image where only local maxima and minima are represented by nonzero values. If the heart is located in the image and the scan line crosses it, some of these local maxima and minima will correspond to anatomical features of the heart. For example, as is shown in Fig. 20, there is a "maximum- minimum-maximum" pattern which has been empirically demonstrated to generally correspond to the long axis of the right ventricle, the interventricular septum, and the long axis of the left ventricle. - In accordance with the preferred embodiment of the invention (which is based upon image data with a typical size of 64 by 64 pixels and prescaled into an intensity range of 0 to 255) , local minima are correlated for connectedness and, where they should be connected, connected with each other. Similarly, local maxima are correlated for connectedness and, where these maxima should be connected, connected with each other. Such correlation is easy to visualize; when a local maximum in one scan line is immediately adjacent a local maximum in the immediately previous line, the two maxima can be taken as part of the same line segment; -if the current local extreme is immediately adjacent the end of a line segment composed of

SUB extremes of the same kind, the current local extreme is connected to the end of the line segment and itself becomes the end of a one-pixel-longer line segment.) Consequently, such correlated local maxima and minima can be grouped into line segments of local maxima and line segments of local minima, respectively. (For computational convenience, these line segments are chosen to be one pixel wide.) To determine which line segments correspond to anatomical landmarks, constraints relating to the anatomical features of interest are established. In the present instance, the system initially "looks" for patterns in which a line segment composed of local minima intervenes along the X coordinate direction between two line segments each composed of local maxima. (Hereinafter, such a pattern will be referred to as an S-pattern.) After finding all such S-patterns, the system then identifies from all of them the single one which best satisfies additional anatomy-specific constraints which are described below.

In a first step, the system scans each S-pattern line by line along the X direction, proceeding from the top down or from the bottom up. Two maximum points and one minimum point are detected by intersections of each scan line with the S-pattern. For each scan line, if these three detected points meet all the following conditions, then the S- pattern is assigned a score of 1 from that scan line, otherwise receiving a score of 0. The conditions are: a) the distance between adjacent maximum points should be approximately 8% to 50% of the width of the image, b) the distance between one minimum point and its adjacent maximum points should be approximately 3% to 25% of the width of the image, c) the intensity of each of the minimum points should be greater than 30% of the intensity of the most intense point in the image, and d) the difference in intensity between a minimum point and its adjacent maximum points should be

S greater than 0.8% of the intensity of the most intense point in the image.

The result of this step is that for each S-pattern, there is a total score which equals the sum of the scores of all the scan lines which intersect it. Then, each of the remaining below-listed constraints is applied to each of the S-patterns. The system then selects the single S- pattern which a) meets all of the below-listed constraints and b) has the highest above-defined score of all the S- patterns which meet all of the below-listed constraints.

Additional constraints:

2) Each line segment of maxima or minima in the S- pattern under consideration should have a maximum average intensity in comparison with that of its corresponding line segment in other S-patterns.

(The average intensity of a line segment is defined to be the sum of the intensities of all pixels along the line segment, divided by the total number of such pixels; in effect, total intensity divided by total length.)

3) Each of the line segments in the S-pattern under consideration should be longer than 5% of the height of the image.

4) Line segments in the S-pattern under consideration should not intersect each other.

5) The positions of the endpoints of the three line segments in the S-pattern under consideration should be consistent.

6) The intensity profile along each of the three line segments in the S-pattern under consideration should be smooth and should not have any deep valleys.

7) The positions, lengths and average intensities of corresponding line segments in the S-pattern under consideration should not change rapidly from one frame to the next. The single S-pattern which satisfies the above

SUBSTIT constraints 2) - 7) and which has the highest score is taken to be the above-referenced "maximum-minimum-maximum" pattern representing the geometry of the heart. In this S- pattern, the rightmost line segment composed of maxima is taken to represent the long axis of the left ventricle, the leftmost line segment composed of maxima is taken to represent the long axis of the right ventricle, and the middle line segment (composed of minima) is taken to represent the interventricular septum. Thus, this step produces three nonintersecting line segments such as are shown in Fig. 6; the leftmost one R generally corresponding to the long axis of the right ventricle, the middle one S generally representing the interventricular septum, and the rightmost one L generally corresponding to the long axis of the left ventricle.

To locate the boundary of the left ventricle in accordance with the preferred embodiment of the invention, a Region of Interest (ROI) is constructed on the basis of the lines corresponding to the interventricular septum and the left ventricle. The definition of the ROI is shown in Fig. 7. As is shown there, the height of the ROI is approximately equal to the height of the septum line or the height of the left ventricle line, whichever is greater. The width of the bottom of the ROI is approximately twice the maximum distance between the detected septum line and the detected left ventricle line. The left edge of the ROI is the septum line. The details of the top, the bottom, and the right boundaries of the ROI are shown in Fig. 7. The thus-defined ROI isolates only the left ventricle. Because the ROI only encompasses a relatively small area and isolates only the left ventricle, it is comparatively easy to locate the boundary of the left ventricle. This is done by using intensity information along the line segment corresponding to the interventricular septum; the intensity along the detected interventricular septum represents the radionuclide activities in heart muscles which immediately adjoin the

SUBST left ventricular blood pool.

In the preferred embodiment, the boundary of the left ventricle is determined adaptively. The maximum intensity along the line segment which corresponds to the interventricular septum is SMAX, the minimum intensity is SMIN, and the length of the line segment is LNT. (Usually, SMAX is located at the top of the line segment and SMIN is located at the bottom of the line segment.) In accordance with the preferred embodiment, an adjustment ratio ADJ is set to equal (SMAX-SMIN)/LNT. Then, the image within the ROI is scanned line-by-line, from the top down. For the Nth scan line an adaptive threshold value is set to equal (SMAX-ADJ*N) , where N ranges from 0 (top of the ROI) to the bottom of the ROI in integral increments. If the intensity of the pixel on the scan line is greater than the current adaptive threshold, the pixel is treated as belonging to the left ventricular blood pool region. Thus, the left ventricular blood pool region within the ROI is adaptively segmented and a one-pixel-wide edge of the blood pool is determined by labeling the boundary of the segmented region.

The above description has concentrated upon the left ventricle because this is the feature which is conventionally monitored in, e.g., ejection fraction computations. While there is no theoretical reason why the boundary of the right ventricle could not be determined in a similar fashion, the task is relatively complicated because there is a substantial overlap between the right ventricle and the right atrium in a 45 degree LAO view and this is usually the position of choice in cardiac studies (because in this position, the left ventricle is separated from the rest of the heart and this is best for cardiac bloodpool studies) .

In further accordance with the invention, it is possible to apply the preferred embodiment of the invention in real time shortly -after the study has been commenced. This is because the preferred embodiment is self-adapted to

SUBSTITUTE different distributions of data, and the anatomical landmarks can be steadily identified both from low statistics data (average of 13 counts per pixel) and high statistics data (average of several hundred counts per pixel) . Another reason why real time application is possible is that the distribution of data (e.g. the locations of intensity maxima and minima) is a function of the structure of the heart and not the duration of the study. Thus, even if there is insufficient data to form an image for diagnosis, it is possible to locate the left and right ventricles and the interventricular septum relative to each other.

The consequence of this is illustrated in Figs. 8 - 17, which are formed using clinical gated SPECT data which were acquired for approximately one real-time minute per camera stop. In Fig. 8, which is taken at approximately 30 degrees LAO, the right ventricle partially overlaps the left ventricle. Fig. 9 shows the anatomical landmarks detected from Fig. 8. In Fig. 10, which is taken at approximately 50 degrees LAO, the right ventricle is partially overlapped by the left ventricle. Fig. 11 shows the corresponding anatomical landmarks detected from Fig. 10. In Fig. 12, which is taken at approximately 40 degrees LAO, the right ventricle is relatively clearly separated from the left ventricle. Fig. 13 shows the corresponding anatomical landmarks detected from Fig. 12.

Consequently, the positioning of the camera with respect to the patient's heart can be deduced from the spatial relationships and parameter relationships (e.g. direction, length, average intensity and shape) between the atrial, ventricular and septal line segments, as will become evident from an examination of Figs. 14 - 17. These figures use the parameter relationships of the detected interventricular septum line as an example, and they relate the parameter of interest to the position of the camera (expressed as the camera stop number) . The detected interventricular septum line is indicated in each of Figs. 9, 11, and 13. If a camera is optimally positioned (such that there is a maximal separation between the left ventricle and the right ventricle) then the average intensity (or event counts) in the interventricular septum line should be at a minimum. This is because the interventricular septum is not overlapped with any portion of either the left or right ventricular blood pools. Additionally, with the camera in the optimal position, the length of the detected interventricular septum should be at a maximum because the septum is fully "visible." Therefore, the position of the point of the detected interventricular septum should be relatively lower. Figs. 14 - 16 show the actual measurements of these three parameters (length, average intensity and end point position vs. camera stop position number) for the above- mentioned SPECT data. Fig. 17 shows how these three parameters correlate with each other, and clearly suggests that camera stop position number 18 (frame 18) is the optimal camera position. This result is consistent with the optimal position determined independently by an experienced clinic technician unaware of the above analysis. Based upon the combination of the above- mentioned spatial relationships between the detected landmarks, the current camera position can be deduced and the optimal camera position can be inferred. This allows the technician to reposition the patient to avoid finishing the study at a suboptimal angle.

Another consequence of this is, in accordance with the preferred embodiment of the invention, the automatic adjustment of camera position to optimize the study which is being conducted. In a conventional motor-driven camera gantry which is capable of performing SPECT studies (such as the ORBITER and DIACAM camera gantries manufactured by Siemens Gammasonics, Inc.) the detector is stepped around the patient's body axis and an image is acquired at each position. If such - a gantry is connected with, and therefore controlled by, a computer programmed in accordance with the present invention, the detector can be set up at an initially preferred position and the study commenced. After minimum sufficient data (rapid scout data) has been accumulated to permit the position of the detector relative to the patient's heart to be determined, the camera may be stepped to a new suggested position determined by the computer in accordance with the present invention. Then, a second set of data can be collected and a new camera position relative to the heart can be determined. The computer can then suggest a final and optimized camera position by comparing the two sets of information which have been accumulated. If necessary, this process can then be repeated until an optimum position has been reached. This is shown in Fig. 18. Positioning of the Patient

Although this aspect of the invention will be described below in connection with a prospective SPECT study of a patient's heart, it will be understood that this is merely an exemplary application. The invention can be adapted to other applications by persons skilled in the art.

Initially, and as is shown in Fig. 1, a cardiac SPECT study is conducted upon a patient 2. In such a study, a radioisotope is administered to the patient 2 and the heart 4 of the patient 2 is imaged using a scintillation camera generally indicated by reference numeral 6. A number of frames of planar images collected by the camera 6 is routed to a computer 8, and the planar image itself may be displayed upon a display 10. Initially, the patient is placed in position and the technician displays, on display 10, a persistence image of the patient's heart to confirm that the positioning is proper (or to detect mispositioning and to correct it) .

In a conventional study, the technician will usually look for particular anatomical landmarks, such as the boundary of the patient's heart. However, if a focussing collimator (such as collimator 11, shown attached to the

SUBSTITUTE SHEET detector of the camera 6) is used, the persistence images of these landmarks may appear distorted and the technician may not be able to recognize them.

Thus, in accordance with the preferred method of the invention (see Fig. 21) , the camera 6 is set up at an initial position with respect to the patient 2 and acquisition of a persistence image (which may be displayed on display 10) is commenced. After a predetermined time (perhaps 1 or 2 minutes, but this is not a part of the invention) the accumulated data in the persistence image is analyzed in accordance with the method and apparatus described previously. The analysis is carried out in such a manner as to computer-identify a set of anatomical landmarks which define the position of the camera with respect to the organ of interest; in the present instance, wherein the heart 4 is the organ of interest, the computer 8 is programmed to computer-identify the muscles of the left ventricle. After such identification, one or more plots of the landmarks of interest (in this instance, the left ventricular muscles) is computer-generated. Then, these plots are displayed and superimposed upon the corresponding locations in the persistence image. This highlights the features of interest. If the feature is mispositioned within the desired Region of Interest, the orientation of the camera 6 may be changed, either manually by the technician (not shown) with or without a suggested new position calculated by the computer 8, or automatically, under the control of the computer 8. After repositioning, the identification, plotting and superimposition steps are repeated for the new position.

The utility of the preferred embodiment of the invention will now be discussed. As can be seen in Fig. 3A, which shows a low-statistics (less than 30 seconds) persistence image on the display 10 that includes the patient's heart 4 and that is formed using a parallel-hole collimator, it can be ^difficult to identify the location of the patient's heart 4 merely by examining the persistence

SUBSTI image. The same point is illustrated in Fig. 4A, which is another low-statistics (less than 30 seconds) persistence image on the display 10 that includes the patient's heart 4 and that is formed using a focussing collimator 11 (in this instance, a fan-beam collimator) .

However, when the method and apparatus for identifying anatomical features, as described above, is used on the image data in Figs. 3A and 4A to computer-identify the position-defining anatomical landmarks (here, the location of the muscles of the left ventricle) of the patient's heart 4, a plot 12 is generated showing those landmarks as identified from data acquired using a parallel-hole collimator and a plot 14 is generated showing that those landmarks as identified from data acquired using a fan-beam collimator. These plots 12 and 14 may then be displayed on the display 10, superimposed on the respective persistence images at the correct locations (see Figs. 3B and 4B) .

Figs. 3B and 4B were not in fact generated using persistence data. In these instances, planar images were used, and the left ventricular cardiac muscles were identified by scanning the image and locating the line segments of local maxima, which were taken to represent these muscles. A line segment of local minima was also located, and was taken to lie along the long axis of the left, ventricle. (The position of the camera with respect to these detected anatomical landmarks can be determined from them because the geometrical relationships between them are already known.) The resulting computer-identified anatomical landmarks were then plotted and the plots were superimposed upon persistence images.

In accordance with the preferred embodiment of the invention, the relative positions of the patient 2 and the detector 6 of the scintillation camera are adjusted by the technician so that the plot is located in a particular predetermined position on the display 10. Thus, for example, the centroid, of the patient's left ventricle may be centered on the display 10, so that when (as in a

SUB myocardial perfusion study) the patient's position is to be replicated in a subsequent study, the positioning may be made highly accurately. Depth Determinations For illustrative purposes, the preferred embodiment of the third aspect of the invention is explained with reference to a cardiac study even though such studies are usually carried out using SPECT. This explanation has been chosen because it utilizes the method and apparatus for identifying anatomical features, but the invention applies equally well to other types of studies.

As is shown in Fig. 1, a nuclear medicine study is conducted upon a patient 2. In such a study, a radioisotope is administered to the patient 2 and the a target organ such as the heart 4 of the patient 2 is imaged using a scintillation camera generally indicated by reference numeral 6. A number of frames of planar images collected by the camera 6 is routed to a computer 8, and the planar image itself may be displayed upon a display 10. It can be seen that the scintillation camera 6 includes a focussing collimator generally indicated by reference numeral 11. The focussing collimator 11 in the illustrated instance is a so-called "fan-beam" collimator with focal length F which magnifies the image in one direction only, but this is not a part of the invention. The focussing collimator 11 may be a collimator which magnifies the image in more than one direction, and need not be of the fan-beam type.

Turning now to Figures 22A and 22B, it can be seen in Fig. 22B, when the patient 2 is closer to the focussing collimator 11, the image of the interventricular septum 14 of the heart 4 as projected upon the sensitive surface of the crystal (not shown) is smaller than when the patient 2 is further away from the sensitive surface of the crystal. The preferred embodiment of the invention utilizes this property to determine-the depth X of the patient's heart 4 (X being defined as the distance between the center of the

SUBST patient's heart 4 and the top of the patient table 18 upon which the patient 2 is placed.)

In accordance with the method and apparatus for identifying anatomical features described above, it is possible to automatically determine anatomical landmarks of interest which relate to a patient's target organ. Thus, the method and apparatus are used to determine the intraventricular septum 14 of the patient's heart 4 as that septum 14 appears in each of the two images. (As is set forth above, this determination can be made even with low- statistics and noisy image data. Thus, even though there may be insufficient data for the two images to be diagnostically useful in and of themselves, there will still be sufficient data to determine the intraventricular septum 14 of the heart 4. )

As can be seen from Fig. 22A, triangle ABC is a right triangle wherein side AB (the length LI of the interventricular septum 14 in the image) forms the shortest side, and side CA has length F. Another triangle DEC is similar to triangle ABC, and by similar triangles:

(1) Ll/F = R/(F-H1+X)

Similarly, as can be seen from Fig. 22B, triangle A'B'C* is a right triangle wherein side A'B' (the projection L2 of the interventricular septum 14 in the image) forms the shortest side, and side CA' has length F.

Another triangle D'E'C is similar to triangle A'B'C, and by similar triangles:

(2) L2/F = R/(F-H2+X)

Therefore, by (1), RF=L1(F-H1+X) and by (2), RF=L2(F- H2+X) . Consequently:

(3) X = ((L1H1-L2H2)/(L1-L2)) - F

It is therefore possible to determine X from two images taken with the same focussing collimator at different heights. Because the method and apparatus for identifying anatomical features is quite robust, it takes only a few minutes to-determine X.

Although the triangles ABC and A'B'C are shown as

SUB right triangles for simplicity of illustration, the above relationships are still valid even when the feature of interest is not located at the center of the field of view and triangles ABC and A'B'C are therefore not right triangles.

Although a preferred embodiment has been described above, the scope of the invention is limited only by the following claims:

Claims

1. A method of identifying, from image data acquired in a nuclear medicine study of a patient's body, an anatomical feature of interest, comprising the steps of: conducting a nuclear medicine study and thereby acquiring an image of the feature, said image including, for each image pixel, an intensity representing that number of detected scintillation events which correspond to the location of that pixel in the patient's body; establishing, on the basis of the feature to be identified, a scan curve and a scan direction; scanning the image with the scan curve along the scan direction and registering, for each location of the scan curve, the pixels with local intensity maxima and minima; grouping together, in accordance with their connectedness, some of such registered pixels, to form maxima line segments composed exclusively of pixels with local intensity maxima and minima line segments composed exclusively of pixels with local intensity minima; and determining which of said line segments correspond with the feature by identifying whether said line segments satisfy constraints which have been empirically determined to relate to the feature.

2. The method of claim 1, wherein the feature is the interventricular septum of the heart, the scan curve is a straight line which is generally normal to the interventricular septum and the scanning direction is generally parallel to the interventricular septum.

3. The method of claim 1, further comprising the step of creating, from the line segments which have been determined to relate to the feature, a region of interest within the image.

4. The method of claim 3, further comprising the step of searching the region of interest for at least one predetermined anatomical feature.

5. The method of claim 4, wherein the feature is the boundary of the left ventricle.

6. A method of identifying, from image data acquired in a nuclear medicine study of a patient's body, an anatomical feature of interest, comprising the steps of: conducting a nuclear medicine study and thereby acquiring an image of the feature, said image including, for each image pixel, an intensity representing that number of detected scintillation events which correspond to the location of that pixel in the patient's body; mapping the image to a local curvature image wherein only pixels with local intensity maxima and minima have non-zero values; grouping together, in accordance with their connectedness, at least some of such non-zero value pixels, to form maxima line segments composed exclusively of pixels with local intensity maxima and minima line segments composed exclusively of pixels with local intensity minima; and determining which of said line segments correspond with the feature by identifying whether said line segments satisfy constraints which have been empirically determined to relate to the feature.

7. A method of optimizing the position of a scintillation camera detector with respect to a target organ within the body of the patient and thereby optimizing image data acquired in a nuclear medicine study, comprising the steps of: acquiring a first image of the target organ at an initially preferred position of the detector with respect to the patient's body,^' said first image including, for each image pixel, an intensity representing that number of 10 detected scintillation events which correspond to the

11 location of that pixel in the patient's body; 2 establishing, on the basis of the target organ, a scan 3 curve and a scan direction;

14 scanning the first image with the scan curve along the 5 scan direction and registering, for each location of the

16 scan curve, the pixels with local intensity maxima and 7 minima;

18 grouping together, in accordance with their L9 connectedness, some of such registered pixels, to form 0 maxima line segments composed exclusively of pixels with 21 local intensity maxima and minima line segments composed -2 exclusively of pixels with local intensity minima; 3 determining which of said line segments correspond 4 with anatomical features of the target organ by identifying 5 whether said line segments satisfy constraints which have 6 been empirically determined to relate to said features; 7 ascertaining, from the line segments which have been 8 so determined to correspond to the feature, a current

29 position of the scintillation camera detector with respect

30 to said target organ; 1 establishing a second preferred position of the 2 scintillation camera detector with respect to the target 3 organ; 4 acquiring a second image of the target organ at said 5 second preferred position of the detector with respect to 6 the patient's body, said second image including, for each 7 image pixel, an intensity representing that number of 8 detected scintillation events which correspond to the 9 location of that pixel in the patient's body; 0 scanning the second image with the scan curve along 1 the scan direction and registering, for each location of 2 the scan curve, the second image pixels with local 3 intensity maxima and minima; 4 grouping together, in accordance with their 5 connectedness, some of such registered second image pixels, to form second image maxima line segments composed exclusively of second image pixels with local intensity maxima and second image minima line segments composed exclusively of second image pixels with local intensity minima; determining which of said second image line segments correspond with anatomical features of the target organ by identifying whether said second image line segments satisfy constraints which have been empirically determined to relate to said features; and ascertaining, from the second image line segments which have been so determined to correspond to the feature, a current position of the scintillation camera detector with respect to said target organ.

8. The method of claim 7, further comprising the step of determining which of the first and second positions is better for conducting the study.

9. The method of claim 7, wherein the scintillation camera detector is moved automatically to an optimized position determined by comparing information acquired at the first position with information acquired at the second position.

10. A method of identifying a patient's cardiac ventricles and interventricular septum from image data acquired in a gated cardiac bloodpool study of the patient's heart, comprising the steps of: acquiring a nuclear medicine image of the heart wherein the interventricular septum extends generally vertically from the top of the image to the bottom of the image, said image including, for each image pixel, an intensity representing that number of detected scintillation events which correspond to the location of that pixel in the patient's heart;

SUBSTITUTE SHEET scanning the image with a horizontally extending line which is moved in a direction that is generally parallel to the patient's interventricular septum and registering, for each vertical location of the horizoritally extending scan line, the pixels with local intensity maxima and minima; grouping together, in accordance with their connectedness, some of such registered pixels, to form maxima line segments composed exclusively of pixels with local intensity maxima and minima line segments composed exclusively of pixels with local intensity minima; identifying S-patterns wherein a minima line segment intervenes, along the horizontally extending scan line, between two maxima line segments; and extracting, from all such identified S-patterns, and in accordance with constraints which have been empirically determined to relate to the interventricular septum and the cardiac ventricles, that single S-pattern which best corresponds to the long axes of the ventricles and the intervening interventricular septum.

11. The method of claim 10, wherein scanning and registration step includes the step of mapping the image to a local curvature image in which only local maxima and minima have non-zero values.

12. The method of claim 10, further comprising the steps of determining, within the image and from the line segments in said single S-pattern, a region of interest which includes only the left ventricle.

13. The method of claim 12, further comprising the steps of adaptively segmenting the left ventricle within the region of interest and creating a graphical plot of the boundary of the left ventricle.

14. The method of claim 13, further comprising the

SUBSTΓΓUTE SHEET step of superimposing the plot upon the image.

15. A method of processing a nuclear medicine image known to contain a feature of interest, comprising the steps of: automatically defining, based on the image itself and without operator intervention, a region of interest within the image; and automatically identifying the feature of interest within said region of interest.

16. A method of identifying anatomical features of interest of a patient's body, comprising the steps of: conducting a nuclear medicine study of the patient's body and thereby acquiring image data; and automatically identifying, based upon the image data and without operator intervention, at least one feature.

17. The method of claim 16, wherein the feature is of a target organ located within a region of interest and wherein the step of identifying includes identifying, with respect to the target organ, at least two anatomical landmarks of interest, the method further comprising the steps of: displaying a persistence image of the region of interest; computer-generating plots representing said at least two anatomical landmarks of interest; and superimposing said plots on corresponding locations in the persistence image.

18. The method of claim 17, further including the step of positioning said plots in a predetermined display position.

19. The method of claim 17, wherein said identifying step comprises the step of identifying said at least two anatomical landmarks from persistence data in a persistence image.

20. The method of claim 16, wherein the feature is of a target organ located within a region of interest; and wherein the step of conducting a nuclear medicine study and acquiring image data includes the steps of: acquiring a first nuclear medicine image of the target organ using a focussing collimator at a first predetermined height; and acquiring a second nuclear medicine image of the target organ using the same collimator at a second predetermined height; and wherein the step of identifying includes the steps of: computer-identifying an anatomical landmark associated with the target organ as said landmark appears in the first nuclear medicine image; computer-identifying the same anatomical landmark associated with the target organ as said landmark appears in the second nuclear medicine image; and determining, from differences in size between said identified landmarks and said heights, the depth of the target organ within the patient.

21. A method of conducting a nuclear medicine study of a target organ within a region of interest within a patient, comprising the steps of: identifying, with respect to the target organ, at least two anatomical landmarks of interest; commencing the study; displaying a persistence image of the region of interest; computer-generating plots representing said at least two anatomical landmarks of interest; and superimposing said plots on corresponding locations in the persistence image.

22. A method of determining the depth of a target organ within a patient, comprising the steps of: acquiring a first nuclear medicine image of.the target organ using a focussing collimator at a first predetermined height; acquiring a second nuclear medicine image of the target organ using the same collimator at a second predetermined height; ^• computer-identifying an anatomical landmark associated with the target organ as said landmark appears in the first nuclear medicine image; computer-identifying the same anatomical landmark associated with the target organ as said landmark appears in the second nuclear medicine image; and determining, from differences in size between said identified landmarks and said heights, the depth of the target organ within the patient.

23. Apparatus for identifying a patient's cardiac ventricles and interventricular septum from image data acquired in a gated cardiac bloodpool study of the patient's heart, comprising: scintillation camera means for acquiring and storing a nuclear medicine image of the heart wherein the interventricular septum extends generally vertically from the top of the image to the bottom of the image, said image including, for each image pixel, an intensity representing that number of detected scintillation events which correspond to the location of that pixel in the patient's heart; and computer means, said computer means comprising: means for scanning the image with a horizontally extending line which is moved in a direction that is generally parallel to the patient's interventricular septum and registering, for each vertical location of the horizontally extending scan line, the pixels with local intensity maxima and minima, means for grouping together, in accordance with their connectedness, some of such registered pixels, to form maxima line segments composed exclusively of pixels with local intensity maxima and minima line segments composed exclusively of pixels with local intensity minima, means for identifying S-patterns wherein a minima line segment intervenes, along the horizontally extending scan line, between two maxima line segments, and means for extracting, from all such identified S- patterns, and in accordance with constraints which have been empirically determined to relate to the interventricular septum and the cardiac ventricles, that single S-pattern which best corresponds to the long axes of the ventricles and the intervening interventricular septum.

24. The apparatus of claim 23, wherein said computer means further includes means for mapping the image to a local curvature image in which only local maxima and minima have non-zero values.

25. The apparatus of claim 23, wherein said computer means further includes means for determining, within the image and from the line segments in said single S-pattern, a region of interest which includes only the left ventricle.

26. The apparatus of claim 25, wherein said computer means further includes means for adaptively segmenting the left ventricle within the region of interest and creating a graphical plot of the boundary of the left ventricle.

27. The apparatus of claim 26, further comprising a display and means for displaying, upon the display, a superposition of the plot and the image.

28. Apparatus for identifying anatomical features of interest in a patient's body, comprising: means for conducting a nuclear medicine study of the patient's body and thereby acquiring image data; and means, responsive to the image data, for identifying at least one feature.

29. The apparatus of claim 28, wherein the feature is of a target organ located within a region of interest and wherein the means for conducting a nuclear medicine study includes means for identifying at least two anatomical features of interest with respect to the target organ, the apparatus further comprising: display means for displaying a nuclear medicine persistence image of the region of interest; and means for generating plots representing said at least two anatomical landmarks of interest and superimposing said plots on corresponding locations in the persistence image.

30. The apparatus of claim 28, wherein the feature is of a target organ and wherein the means for conducting a nuclear medicine study includes means for acquiring and storing a first nuclear medicine image of the target organ using a focussing collimator at a first predetermined height and acquiring and storing a second nuclear medicine image of the target organ using the same focussing collimator at a second predetermined height; and wherein the means responsive to the image data includes means for identifying an anatomical landmark associated with the target organ as said landmark appears in the first nuclear medicine image, for identifying the same anatomical landmark associated with the target organ as said landmark appears in the second nuclear medicine image, and for determining, from differences in size between said identified landmarks and said heights, the depth of the target organ within the patient's body.

31. Apparatus for conducting a nuclear medicine study of a target organ within a region of interest within a patient, comprising: a computerized scintillation camera system for acquiring and storing nuclear medicine images of a patient and for automatically identifying at least two anatomical features of interest with respect to the target organ; display means for displaying a nuclear medicine persistence image of the region of interest; and means for automatically generating plots representing said at least two anatomical landmarks of interest and superimposing said plots on corresponding locations in the persistence image.

32. Apparatus for determining the depth of a target organ within a patient, comprising: scintillation camera means for acquiring and storing a first nuclear medicine image of the target organ using a focussing collimator at a first predetermined height and acquiring and storing a second nuclear medicine image of the target organ using the same focussing collimator at a second predetermined height; and computer means for identifying an anatomical landmark associated with the target organ as said landmark appears in the first nuclear medicine image, for identifying the same anatomical landmark associated with the target organ as said landmark appears in the second nuclear medicine image, and for determining, from differences in size between said identified landmarks and said heights, the depth of the target organ within the patient.