Low signal correction for perfusion measurements

[0002] The present invention relates to the art of medical diagnostic imaging. It finds particular application in conjunction with calculating tissue perfusion using computed tomography (CT) scanners, and will be described with particular reference thereto. However, it is to be appreciated that the present invention is also amenable to other modalities such as MRI, and is not limited to the aforementioned application.

[0003] Generally, CT scanners have a defined examination region or scan circle in which a patient, or subject being imaged is disposed on a patient couch. A fan beam of radiation is transmitted across the examination region from an radiation source, such as an x-ray tube, to an oppositely disposed array of radiation detectors. The x-ray tube and associated power supply and cooling components are rotated around the examination region while data is collected from the radiation detectors. Rotation of the radiation source is often achieved by mounting the radiation source to a rotating gantry which is rotated on a stationary gantry. For volume imaging, the patient couch is moved longitudinally. Continuous movement achieves spiral scanning whereas discrete steps achieve a series of parallel slices.

[0004] The sampled data is typically manipulated via appropriate reconstruction processors to generate an image representation of the subject which is displayed in a human-viewable form. Various hardware geometries have been utilized in this process. In third generation scanners, both the source and detectors rotate around the subject. In a fourth generation scanner, the x-ray source rotates and the detectors remain stationary. The detector array typically extends 360° around the subject in a ring outside of the trajectory of the x-ray tube.

[0005] In a perfusion study, blood flow in tissues and vessels of interest is of primary concern. Typically, a contrast agent is injected into the subject and multiple “snapshots” of the region of interest are taken over time. Present CT scanners are capable of taking 1 to 2 snapshots per second of the region, providing a series of images that tracks the contrast agent in near-real time.

[0006] One particular application of CT perfusion is helping to diagnose cerebral ischemia in patients who have suffered acute strokes. This type of study requires precise measurements over a period of time. One technique that is used in the calculation of perfusion is the maximum slope method, which calculates the maximum slope of a time vs. density curve and a maximum arterial enhancement. Perfusion is the maximum slope divided by the maximum arterial enhancement. Accuracy of the quantitative data is impacted by noise in the data, which may have several possible sources. These include patient motion, blood recirculation, partial volume effect, and other factors.

[0007] One method of reducing patient motion in a head CT scan, and thus improving the quality of the perfusion investigation, is immobilizing the head of the subject in an external restraint. Typically, such a device includes a strap that is connected to the patient couch that traverses the forehead of the subject, effectively eliminating head motion in a vertical direction (given that the subject is laying horizontally). However, the subject is still capable of movement laterally, as well as slight rotation of the head. These movements can seriously degrade the quality of a perfusion study, causing misalignment of the series of images, blurring a resultant image, and having adverse effects on the calculation of blood perfusion. The maximum density enhancement, measured in Hounsfield units (HU) can be reduced by 40% or more by motion that can occur despite the aid of a head restraint. The blurred images, and effects on perfusion measurements significantly impact the accuracy of quantitative measurements used in diagnosis.

[0008] Further, background noise is a factor that affects perfusion calculation, as well as the images associated therewith. Regions that exhibit low signal can be overshadowed by noise. In low blood flow regions, the maximum density enhancement and the noise can both be in the 2-4 Hounsfield unit range. Legitimate perfusion signals can be hidden decreasing the efficacy of the study as a whole. Filters meant to eliminate noise may also eliminate low strength perfusion signals effectively getting rid of good information along with useless information.

[0009] The present invention contemplates a new and improved method and apparatus which overcome the above-referenced problems and others.

[0010] In accordance with one aspect of the present invention, a method of correcting for low signal is provided. A volume image of a region of interest is generated. Low signal voxels are identified and grouped into clusters. Perfusion values are determined for grouped and ungrouped voxels, and a single profusion image is generated.

[0011] In accordance with another aspect of the present invention, a method of improving a signal-to-noise ratio in provided. A perfusion study is performed, and time-density curves are generated for all voxels of a perfusion image. Low signal voxels are identified and grouped together. Representative time-density curves are found for each low signal voxel group, and function values are calculated therefrom. The results are applied to each member voxel of the group. The results are combined with results from normal signal voxels to produce a single perfusion image.

[0012] In accordance with another aspect of the present invention, a diagnostic imaging apparatus is provided. A signal analyzer monitors time-density curves of voxels of a volume image. A voxel binner groups low signal voxels. A voxel combiner combines intensity values of the voxels within the group. A perfusion calculator calculates perfusion values for the low and normal signal voxels.

[0013] One advantage of the present invention is a reduction of the negative effects of patient motion.

[0014] Another advantage resides in a reduction of the partial volume effect.

[0015] Another advantage resides in the reduction of the negative effects of blood recirculation.

[0016] Another advantage resides in the reduction of the effect of low amplitude signals.

[0017] Another advantage resides in the increased accuracy of curve fits.

[0018] Another advantage resides in reduction of errors caused by noise.

[0019] Still further advantages and benefits of the present invention will become apparent to those of ordinary skill in the art upon reading and understanding the following detailed description of the preferred embodiments.

[0020] The invention may take form in various components and arrangements of components, and in various steps and arrangements of steps. The drawings are only for purposes of illustrating preferred embodiments and are not to be construed as limiting the invention.

[0021] FIG. 1 is a diagrammatic illustration of a computed tomography scanner in accordance with the present invention;

[0022] FIG. 2 is a flow diagram that includes integral steps of the present invention.

[0023] With reference to FIG. 1, a CT scanner 10 includes a stationary gantry 12 and a rotating gantry 14 which define an imaging region 16. The rotating gantry 14 is suspended from the stationary gantry 12 for rotation about the examination region 16. A radiation source 20, such as an x-ray tube, is arranged on the rotating gantry 14 for rotation therewith. The radiation source 20 produces a beam of penetrating radiation that passes through the examination region 16 as the rotating gantry 14 is rotated by an external motor (not illustrated) about a longitudinal axis of the examination region 16. A collimator and shutter assembly 22 forms the beam of penetrating radiation into a cone shape and selectively gates the beam on and off. Alternately, the radiation beam is gated on and off electronically at the source 20. A subject support 30, such as a radiolucent couch or the like, suspends or otherwise holds a subject being examined or imaged at least partially within the examination region 16 such that the cone-shaped beam of radiation defines a volume through the region of interest of the subject.

[0024] The imaged volume is repeatedly imaged over a period of time. In a perfusion study, a contrast agent is injected into the subject and factors relating to blood flow of the subject are monitored over a period of time to track blood flow behavior in the region of interest. The volume is segmented into a three dimensional array of voxels, which are often conceptualized as a series of slices, each slice having a finite thickness.

[0025] In a third generation CT scanner, an array of radiation detectors 40 is mounted peripherally across from the source on the rotating gantry. In a fourth generation CT scanner, a stationary ring of radiation detectors 42 is mounted around the stationary gantry 12. Regardless of the configuration, the radiation detectors are arranged to receive the radiation emitted from the source 20 after it has traversed the imaging region 16.

[0026] The radiation detectors 40, 42 convert the detected radiation into electronic projection data. That is, each of the radiation detectors produces an output signal which is proportional to an intensity of received radiation. Each radiation detector generates data elements which correspond to projections along a corresponding ray within the view. Each element of data in a projection or data line is related to a line integral of an attenuation coefficient taken along its corresponding ray passing through the subject being reconstructed.

[0027] A data memory or buffer 50 receives the sampled data from the radiation detectors. The data memory 50 optionally performs filtering or other operations before passing the data to a reconstruction processor 52 which reconstructs volume image representations of the subject.

[0028] In the preferred embodiment, the gantry 14 makes approximately 40 turns around the subject, to produce 40 volume images of the region of interest which are stored in a first series of image memories. of course, the number of images can be more or less, 40 is a balance between factors such as time of scan, radiation dose to the subject, cardiac cycle, and a period of time wherein useful perfusion information can be gathered. Typical present day CT scanners can generate 40 images in about 20-40 seconds, which is a relatively long time that the subject is asked to remain perfectly motionless. In order to correct for inevitable patient motion, a registration processor analyzes the volume images and aligns them such that the region of interest remains stationary over the course of the images.

[0029] The registration processor selects a corresponding reference slice in each of the 40 volume images which it actively calculates a movement function. The reference slice is preferably a central slice. In the preferred embodiment, a diagnostician is presented (on a user input terminal) with an image of the reference slice. This first image of the reference slice is used as the norm to which each subsequent or preceding time-step image is compared and adjusted to match.

[0030] Preferably, the registration processor identifies landmarks which are easy to identify, shapely defined and appear in diverse parts of the slice. In a brain perfusion scan, an exemplary landmark is a portion of the skull, having constant shape and intensity from image to image over the whole scan period. Optionally, the diagnostician can crop the slice to a subregion of interest to reduce processing time.

[0031] Each subsequent image of the reference slice is searched in this manner for the selected region, and each subsequent image is shifted or rotated to bring the landmarks into alignment with the reference image. As the registration processor aligns these images, it records a movement function for each of the 40 volume images that describes its movement relative to the reference slice. Especially in a head scan, the region of interest can be considered a rigid body, and any movement that the reference slice undergoes, the entire imaging volume undergoes. The recorded movement function is applied to each slice of the corresponding volume image to align the remainder of the imaging volume. Alternately, and more time intensive, the alignment process can be performed individually for each slice of each volume. Other alignment processes and algorithms are also contemplated.

[0032] Some voxels within the region of interest have weak time-density curves. More specifically, some voxels have amplitudes that are comparable with noise. The preferred embodiment groups similar weak signals and combines them to make characteristic stronger signals.

[0033] After the reconstruction processor 52 has reconstructed the volume images of the region of interest, the volume images are divided into slices and stored in a high resolution slice image memory 60. The slice image memory includes n submemories 60₁, 60₂, . . . 60_n, where n is the number of slices in the imaging volume. That is, the first slice of the 40 temporally displaced volume images from the beginning of the scan to the end of the scan are stored in order in a first slice submemory 60₁, the images of the second slice are stored in a second slice submemory 60₂, and so on to the images of the n^thslice which are stored in an n^thsubmemory 60_n. In the preferred embodiment, the slices are each one voxel thick. In the preferred embodiment, a 512×512 image matrix is used, and each slice is one voxel thick. That is, the 40 density values from each corresponding voxel in the 40 slices define a time vs. density curve. The slices are sorted 62 by voxel and a time-density curve calculator 64 calculates a time density curve for each voxel. Therefore, in the preferred embodiment in which each slice is 512×512, there are 512×512 time-density curves per slice. A curve for each voxel of each slice stored in a curve memory 66.

[0034] The intensity values of the corresponding voxels of the 40 volume images define a time density curve. Each time density curve is a measure of the amount of contrast agent within the subregion corresponding to the same voxel in each of the time displaced volume images. A typical time-density curve includes a leading edge during which the contrast agent is entering the voxel region rapidly, a maximum at which time the contrast agent is at a maximum concentration, and a trailing edge during which the contrast agent is leaving the voxel. The curve typically is a gamma-variate curve which is characterized by its steep leading edge and gradual trailing edge.

[0035] A maximum enhancement processor searches for the maximum enhancement value of the time-density curves of the voxels within an artery region indicated by the diagnostician on the reference slice. More specifically, the maximum intensity processor searches for the maximum enhancement among all voxels in a diagnostician indicated artery region. The maximum enhancement of the artery is used later in a perfusion calculation.

[0036] The high resolution slices are passed through a filter and subsequently reduced in resolution by a resolution reducer. The resolution reducer takes a high resolution image matrix of each slice in time, groups the voxels, and combines each group of voxels, e.g. averages, maximum intensity, etc. In the preferred embodiment, the high resolution matrices are 512×512, and the low resolution matrices are 128×128. The resolution reducer bins the voxels into groups of 16 by position, that is, 4×4 groups of high resolution voxels are combined into a single low resolution voxel. After the volume images over the whole scan time are reduced in resolution, they are stored in a low resolution memory.

[0037] The low resolution images are used to calculate a number of factors that are later used in the perfusion calculation. More specifically, a low signal filter eliminates low signals. The low signal filter identifies the voxels that have time-density curves too weak or too poorly defined to be used by themselves. At least one of multiple criteria is used to determine which signals are too weak. One method is to compare the time-density curve to a curve model. Voxels having curves outside of a preselected range of fit to the model are discarded as having low signal. Another method is to find a peak enhancement value of the time density curve for each voxel. Voxels with peak enhancements lower than a preselected threshold enhancement value are discarded as having low signal. Another method of identifying low signal voxels is selecting voxels that are historically of low signal, e.g. bone.

[0038] Typically, the patient's circulatory system recirculates the contrast agent back through the region of interest causing a secondary intensity peak. If the secondary peak is included in the gamma-variate curve fitting, the peak is shifted later in time altering the slope of the leading edge. A clipping circuit clips the secondary peak based on percentage intensity drop from the maximum, a time after the maximum, or a combination of the two. A processor replaces the clipped region with a gamma variate curve segment or other extrapolation of the remaining curve portion. A curve fitting processor compares the time-density curves to a model curve. Data that is not within a preselected tolerance of the ideal curve is filtered out as bad data.

[0039] More specifically, a gamma-variate curve smoothing circuit smooths the time density curve of each voxel to reduce noise. The smoothed curves are mathematically fit to a gamma-variate curve. More specifically, the value K, value α, and value β that define a gamma variate curve mathematically are calculated. Voxels that have a better fit to the gamma-variate model typically have a stronger signal, and are thus more robust for use in the perfusion calculation. A maximum slope calculator calculates the maximum slope of the region of the time-density curve from the K, α, and β values.

[0040] A blood perfusion value is now calculated for each voxel. In a preferred embodiment a perfusion calculator 110 divides the maximum slope value for each voxel by the maximum artery enhancement found the maximum enhancement processor to obtain a perfusion value for each voxel. More specifically, a high signal perfusion calculator 110_acalculates perfusion for all non-grouped time-density curves, and a low signal perfusion calculator 110_bcalculates perfusion for all grouped time-density curves. An interpolator interpolates the truncated time-density curve to form representative curves. Alternately, the K, α, β, and maximum enhancement values can address a preloaded look-up table to retrieve the perfusion value. These values are stored in a perfusion image memory 112. A video processor 114 places data from the perfusion image memory 112 in proper format for a video monitor 116.

[0041] Voxels identified as having low amplitude time density curves are identified and sorted by a processor 120. Optionally, the low amplitude data is temporally filtered to eliminate curves that are not generally contemporaneous to the curves of neighboring voxels. A sorter 124 sorts the time-density curves into groups. Each group is averaged, or summed, or otherwise combined by a curve averaging processor 126 and the combined time density curve replaces the time density curve of all curves in the group. The sorter 124, in the preferred embodiment, groups the voxels using one of k-means clustering, c-means clustering, and fuzzy logic. It is to be understood that other methods of grouping voxels may also be utilized. The curve averaging processor groups voxels with similar characteristics together. The voxels are determined to be similar based on at least one of its x-coordinate position, its y-coordinate position, its peak enhancement value, a time the time density curve takes to reach the peak enhancement value (time-to-peak), the Hounsfield number, and the like. Once the time density curves are grouped and combined, the groups are passed to the curve fitting processor. This greatly reduces the inherent noise in the signals as the noise tends to cancel out as the signals are averaged. Thus, the averaged signal has a higher signal-to-noise ratio than the individual curves of any of the constituent voxels of the group. The curve fitting processor fits the combined time-density curve and fits it to the model curve. A common perfusion value is determined for all the constituent voxels of the group.

[0042] For example, the voxels are grouped by x and y-coordinate positions. This scheme yields voxel groups containing constituent voxels that are physically close to each other. In another example, voxels are grouped solely by maximum Hounsfield number, voxels with maximum Hounsfield values of 2-4 HU are grouped together, and voxels with values of 4-8 HU are grouped together, regardless of spatial position. Preferably, the combination of criteria that best serves each individual perfusion study is selected. In this manner, one perfusion image is made from the normal signal voxels and the plurality of low-signal voxel groups.

[0043] With reference to FIG. 2, a preferred embodiment is presented in a flowchart form. In a perfusion study 130 time-density curves of voxels of an imaging region are computed. In a low signal identification step 132, all voxels with low signal are identified. In a clustering step 134, low signal voxels are clustered together. In a representative determination step 136 representative time-density curves are computed. In a functional measurement step 138, measurements are calculated from the combined and uncombined time-density values. In an assigning step 140, each low signal voxel is assigned the values determined for its group. In a combining step 142 the results of the low and normal signal voxels are combined to produce a single functional perfusion image.

[0044] The invention has been described with reference to the preferred embodiments. Obviously, modifications and alterations will occur to others upon reading and understanding the preceding detailed description. It is intended that the invention be construed as including all such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.