JP2007260292A - Image processor and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To highly reliably evaluate the noise of a medical image even when noise components are not uniform. <P>SOLUTION: An image processor includes: a storage part 22 for storing data for the medical image; a component separation processing part 26 for estimating a noise component distribution by allowing a signal component distribution to be expressed by a polynomial expression to be approximated with respect to an observation signal distribution concerning the partial area of the medical image; and an SNR calculating part 27 for calculating a noise index concerning the partial area being the same as the former partial area or the partial area of another position, based on the estimated noise component distribution. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to an image processing apparatus and program for performing noise evaluation on a medical image.

  There are several generally known methods for calculating a signal-to-noise ratio (SNR) and a contrast-to-noise ratio (CNR) of an image. In terms of measuring noise components, the following three are given as typical examples.

(1) Difference method
By taking the difference between images captured under the same conditions and removing reproducible components (including Gibbs ringing in addition to signals), components corresponding to random noise are extracted. This standard deviation (standard deviation SD) is measured. This is an appropriate method for phantoms. However, it is a problem that repeated imaging under the same conditions is difficult when a person is targeted, especially in the case of clinical cases.

(2) Method without difference: measured as it is at the region of interest (also called “same region of interest method” etc.)
The signal component is required to be flat, but MR images often have low-order spatial components (slow signal changes) due to various factors. If a person is a subject, the original anatomical structure is inevitably present, so it is often not satisfied. This is also difficult to solve for humans, and this method is not an appropriate method.

(3) Method without difference: Substitute with background SD (also called “air noise method”)
The signal component is measured in the region of interest ROI, but the noise component is substituted with the standard deviation SD or average value of the background portion where there is no signal. Since the noise characteristics are different in the background portion of the absolute value image, the measured value in the background is converted into the noise amplitude of the signaled portion. This is the most widely used method. In images obtained with a single coil or images obtained by the Sum of Square method (SOS method, or more precisely, the Square Root of Sum-of-Square method), which is a typical image generation method for array coils, the problem is There wasn't much.

  For example, as shown in FIG. 9, the standard deviation σ (B) is obtained assuming that the background part B is a noise distribution, and the signal part converted value that changes depending on the number of channels of the array coil from the standard deviation σ (B). By obtaining σ ′ = σ ′ (B), noise representing this image can be obtained. The SNR in the real organ 1 may be the average value m (O1) of the observation signal in the local region divided by the σ ′, that is, m (O1) / σ ′. Similarly, the CNR between the parenchymal organs 1 and 2 is calculated as (m (O1) −m (O2)) / σ ′. Clinically, the CNR of the parenchymal organ adjacent to the affected area L is expressed as (m (L) −m (O1)) / σ ′.

Both methods are considered on the assumption that the noise intensity does not fluctuate relatively in the image and is uniform. Therefore, when the noise intensity relatively fluctuates depending on the position in the image, there is a problem that the reliability of the conventional noise evaluation is lowered. For example, in recent years, with the progress of parallel imaging (PI) in MRI, noise is not spatially uniform due to the sensitivity unevenness correction processing of the surface array coil used therefor and the PI expansion processing. Therefore, there is no appropriate method for estimating the noise component.
National Electrical Manufacturers Association: Determination of signal-to-noise ratio in diagnostic magnetic resonance imagers, NEMA Standard Publications, MS-1, 1988 Kasai, Doi, MR Imaging Technology, Ohmsha, 2001 Ogura, et al .: Basic study on SNR measurement of MR images, Journal of Japanese Society of Radiological Technology, 59 (4), 508-513, 2003 Pruessmann KS, et al, SENSE: Sensitivity Encoding for Fast MRI, Magnetic Resonance in Medicine 42: 952-962 (1999)

  An object of the present invention is to perform noise evaluation of a medical image with high reliability even when noise components are not uniform.

  In one aspect, the present invention estimates a noise component distribution by approximating a signal component distribution expressed by a polynomial to a storage unit that stores medical image data and an observed signal distribution related to a partial region of the medical image. An estimation unit; and a noise index calculation unit that calculates a noise index related to a partial region that is the same as or different from the partial region based on the estimated noise component distribution.

  According to the present invention, it is possible to perform noise evaluation of a medical image with high reliability even when noise components are not uniform.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. Note that the present invention is an image processing apparatus and its program for noise evaluation of medical images, and the medical image to be targeted in the present invention has a noise intensity that appears non-uniformly. Is an MRI image obtained by parallel imaging using a magnetic resonance imaging apparatus, but is a projection image obtained by another X-ray diagnostic apparatus, an ultrasonic image such as a B mode by an ultrasonic diagnostic apparatus, and a spatial distribution of radioisotopes by a gamma camera. In addition, a tomographic image obtained by an X-ray computed tomography apparatus may be a target. Here, an MRI image will be described as an example of a medical image. Although the image processing apparatus is described as an external apparatus connected to the magnetic resonance imaging apparatus here, a configuration in which the image processing apparatus is incorporated into the magnetic resonance imaging apparatus may be used.

  FIG. 1 shows the configuration of an image processing apparatus according to this embodiment together with a magnetic resonance imaging apparatus. The magnetic resonance imaging apparatus has a cylindrical static magnetic field magnet 1 that forms a static magnetic field H0. A gradient magnetic field coil unit 2 and an RF coil unit 3 are arranged inside the static magnetic field magnet 1. The gradient coil unit 2 has three coils corresponding to the XYZ axes. The gradient magnetic field power supply 4 is configured to individually supply current to the three coils of the gradient magnetic field coil unit 2 according to the control of the gradient magnetic field power supply control unit 7. The RF coil unit 3 includes a plurality of surface coils, and receives a high-frequency signal from the RF generator 5 according to the control of the RF control unit 8 to generate a high-frequency magnetic field. The reception detector 6 receives the MR signal from the nuclear spin inside the subject P through the RF coil unit 3 under the control of the receiver controller 9. The sequence controller 10 controls the gradient magnetic field power supply control unit 7, the RF control unit 8, and the reception device control unit 9 in order to execute a pulse sequence constructed for parallel imaging. The computing unit (reconstruction) 11 reconstructs the MR image according to the processing procedure corresponding to the parallel imaging based on the MR signal received by the reception detector 6 through the plurality of surface coils of the RF coil unit 3.

  The image processing apparatus 20 inputs MR image data via the interface 29 and stores it in the memory 22. Along with this memory 22, the image processing apparatus 20 includes a CPU 21 responsible for overall operation control and various arithmetic processes, an input device 23 such as a keyboard and a mouse, a monitor for displaying MR images, a noise component SI (N) described later, and the like. 24. Further, the signal intensity SI (O) of the observation signal in the partial region of the MR image (hereinafter referred to as the region of interest ROI) is changed into the signal intensity SI (S) of the signal component and the signal intensity SI (N) of the noise component. A standard deviation σ or SNR (SNR) is calculated as a noise index for the region of interest ROI from the component decomposition processing unit 26 for decomposition, the signal intensity SI (O) of the observation signal, and the signal intensity SI (N) of the noise component. SNR calculator 27 and display processor 28 responsible for generating a display image or the like from noise component SI (N). In the following description, “signal strength” is abbreviated.

  FIG. 2A shows a noise evaluation processing procedure according to this embodiment. First. As shown in FIG. 3, at step S <b> 11, at least one region of interest ROI is set on the MR image displayed on the monitor 24 via the input device 23. As shown in FIG. 4, the observation signal SI (O) is composed of a narrowly defined signal component SI (S) and noise component SI (N). Typically, the signal value of the signal component SI (S) changes gently in space, and the signal value of the noise component SI (N) changes randomly in space.

  In step S12, the component decomposition processing unit 26 decomposes the observation signal SI (O) in the region of interest ROI into a signal component SI (S) and a noise component SI (N). In practice, the spatial distribution of the signal component SI (S) represented by a polynomial model (surface model) is typically used as an approximation for the spatial distribution of the observation signal SI (O) in the region of interest ROI. Typically, approximation is performed by the method of least squares. An approximation error is estimated as the noise component SI (N).

  More preferably, the approximate model is selectively used according to the feature of the signal (anatomical structure) in the region of interest ROI. Depending on the object, a least square approximation process using a linear function may be used, but usually a least square approximation process using a polynomial of quadratic or higher is performed. In other words, least square approximation processing, in other words, regression analysis processing is performed on the spatial distribution of the observation signal, and its error component (residual component) is estimated and used as a noise component.

  Particularly in clinical images, it is difficult to set the region of interest ROI, and statistically, it is desired to secure a larger number of points, but because of the anatomical structure, it is necessary to make it relatively small. When the signal component in the region of interest ROI seems to have a certain degree of uniformity, an approximation process using a linear function (linear function) is often sufficient. This corresponds to a case where the white matter portion can be secured in a wide range in the liver or spleen under imaging conditions where contrast is not so great, or the top of the head. In gray matter, it is difficult to take a large area, and in some cases, the least square approximation using a second-order polynomial may be meaningful. However, it is necessary to confirm well whether approximation is successful in a polynomial of a second or higher order.

  In step S13, the SNR calculation unit 27 calculates the standard deviation σ related to the noise component SI (N) in the region of interest ROI estimated by the component decomposition processing unit 26 as a noise index related to the region of interest ROI. When a plurality of regions of interest ROI are set, an approximation process and a standard deviation σ calculation process are individually performed. Further, the display processing unit 28 three-dimensionally associates a noise level as shown in FIG. 5A with a height from the spatial distribution of the noise component SI (N) in the region of interest ROI in accordance with an instruction from the operator. A map to be expressed, a noise level as shown in FIG. 5B on the vertical axis, a map in which the horizontal axis is associated with the X or Y space axis and expressed two-dimensionally, and a noise level as shown in FIG. 5C is on the horizontal axis The histogram is created by any one or any combination of histograms in which frequencies are associated with the vertical axis. The noise component map and histogram created by the display processing unit 28 are displayed on the monitor 24 together with the standard deviation σ as a noise index. In consideration of variations in signal variation within each ROI of the clinical image, the noise situation is confirmed, and this display can prevent an error in image evaluation.

  In the present embodiment, the noise evaluation process may be performed according to the procedure shown in FIG. 2B. Steps S21 and S22 correspond to steps S11 and S12 of FIG. In step S23, the SNR calculation unit 27 calculates the standard deviation σ related to the noise component SI (N) in the region of interest ROI estimated by the component decomposition processing unit 26 as a noise index related to the region of interest ROI. In step S24, the average value m of the observation signal SI (O) of the region of interest is divided by the standard deviation σ to calculate the SNR as a noise index related to the region of interest ROI. Note that when a plurality of regions of interest ROI are set, the standard deviation σ calculated in a specific region of interest ROI may be shared with the SNR calculation of other regions of interest ROI as a unique noise value of the image. Alternatively, the approximation process and the standard deviation σ calculation process may be individually performed for a plurality of regions of interest ROI.

  In the present embodiment, the noise index may be relatively evaluated between two regions of interest by the procedure shown in FIG. 2C. Steps S31 and S32 correspond to steps S11 and S12 in FIG. 2A, respectively, and step S33 corresponds to step S23 in FIG. In step S34, the SNR calculation unit 27 is a representative value of the standard deviation σ01 related to the first region of interest ROI01 and the standard deviation σ02 related to the second region of interest ROI02 estimated by the component decomposition processing unit 26. The value obtained by dividing the difference (m01−m02) between the average value m01 of the observation signal SI (O) of the region of interest ROI and the average value m02 of the observation signal SI (O) of the second region of interest ROI is a relative noise index. Calculated as CNR. As the representative values of the standard deviation σ01 and the standard deviation σ02, typically, a high value is selected. That is, when σ01> σ02, σ01 is selected. However, the average value ((σ01 + σ02) / 2) of the standard deviation σ01 and the standard deviation σ02 may be used as the representative value.

  In the present embodiment, the approximation process may be performed by excluding abnormal values by the procedure shown in FIG. FIG. 8A shows an example in which a blood vessel having a high signal intensity is included in the ROI or an abnormal value is shown in a spike shape on the image. In the designation of the actual region of interest ROI on the clinical image, there may be a case where large noise is mixed due to a factor other than random noise, or a blood vessel signal or the like enters the ROI. Specifying an appropriate threshold value, excluding such “abnormal points”, and performing the approximation process again is effective in improving the accuracy of the approximation process. Step S41 corresponds to step S11 in FIG. In step S42, it is determined for each point in the region of interest ROI whether or not it is an abnormal value. In step S43, the component decomposition processing unit 26 decomposes the observation signal SI (O) in the region of interest ROI into a signal component SI (S) and a noise component SI (N). In step S44, the SNR calculator 27 calculates a standard deviation σ related to the noise component SI (N) in the region of interest ROI estimated by the component decomposition processor 26. Subsequently, in the component decomposition processing unit 26, an error component of each point (each pixel) in the region of interest ROI, that is, the noise component SI (N) is set as a predetermined threshold value according to the standard deviation, for example (3 · (σ) (S45). Then, the component decomposition processing unit 26 extracts points where the noise component SI (N) is larger than the threshold, excludes these points from the approximation processing target (S46), and targets the points remaining in the region of interest ROI. The approximation process is performed again in step S43. In this way, the loop of steps S43 to S46 is repeated until there are no abnormal points or the number of abnormal points converges below a predetermined number. When the abnormal point disappears or the number of abnormal points converges below a predetermined number, the SNR and CNR are calculated from the standard deviation σ at that time and the average value m of points other than the abnormal point (S47).

  In the example of FIG. 6, the abnormal value exclusion is automatically performed. However, as shown in the procedure of FIG. 7, the operator may determine whether to exclude the abnormal value. In step S42, it is determined for each point in the region of interest ROI whether or not it is an abnormal value. In steps S51 to S54, similar to steps S41 to S44 in FIG. 6, an approximation process is performed for each region of interest ROI to obtain the spatial distribution of the noise component SI (N), and the standard deviation σ is calculated. Subsequently, in step S55, the display processing unit 28 associates the noise level as shown in FIG. 8A with the height from the spatial distribution of the noise component SI (N) in the region of interest ROI according to the operator's instruction. A map expressed three-dimensionally, a noise level as shown in FIG. 8B, a map expressing two-dimensionally with the vertical axis representing the noise level as shown in FIG. Is generated by any one or any combination of histograms in which the horizontal axis represents the frequency and the vertical axis represents the frequency, and is displayed on the monitor 24. In these maps and histograms, points as exclusion candidates that typically exceed (3 · σ) as a predetermined threshold are displayed in a unique manner such as red. The operator confirms the map and the histogram, and designates whether or not to exclude the abnormal value from the approximation process via the input device 23 (S56).

  When the abnormal value is excluded from the approximation process, the component decomposition processing unit 26 excludes the point where the noise component SI (N) is larger than the threshold value (S57) and repeats the step for the point remaining in the region of interest ROI. In S53, approximation processing is performed. Thus, the loop of steps S53 to S57 is repeated until the operator gives an instruction not to exclude the abnormal value from the approximation process as the number of abnormal points decreases. When the operator gives an instruction not to exclude abnormal values from the approximation process, SNR and CNR are calculated from the standard deviation σ at that time and the average value m of points other than the abnormal points (S58).

  According to the present embodiment, even when there is no spatial uniformity of noise, it is possible to appropriately estimate the degree of noise in each part. Furthermore, in clinical images, it is possible to more robustly estimate the SNR at a region of interest and the CNR between regions of interest than before.

  Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the components without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of components disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different embodiments may be appropriately combined.

1 is a diagram showing a configuration of a magnetic resonance imaging apparatus equipped with an image processing apparatus according to an embodiment of the present invention. The flowchart which shows the 1st noise evaluation processing procedure by the image processing apparatus of FIG. The flowchart which shows the 2nd noise evaluation processing procedure by the image processing apparatus of FIG. The flowchart which shows the 3rd noise evaluation processing procedure by the image processing apparatus of FIG. The figure which shows the region of interest ROI set by S11 of FIG. 2A, and its observation signal distribution. Supplementary explanatory drawing of S12 of FIG. 2A. The figure which shows the 1st example of a display in S13 of FIG. 2A. The figure which shows the 2nd display example in S13 of FIG. 2A. The figure which shows the 3rd example of a display in S13 of FIG. 2A. The flowchart which shows the 4th noise evaluation processing procedure by the image processing apparatus of FIG. The flowchart which shows the 5th noise evaluation processing procedure by the image processing apparatus of FIG. The figure which shows the 1st example of a display in S28 of FIG. The figure which shows the 2nd example of a display in S28 of FIG. The figure which shows the 3rd display example in S28 of FIG. Supplementary explanatory drawing regarding conventional noise evaluation.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Static magnetic field magnet, 2 ... Gradient magnetic field coil unit, 3 ... RF coil, 4 ... Gradient magnetic field power supply, 5 ... RF generator, 6 ... Reception detection apparatus, 7 ... Gradient magnetic field power supply control part, 8 ... RF control part, DESCRIPTION OF SYMBOLS 9 ... Reception apparatus control part, 10 ... Sequence controller, 11 ... Operation unit, 20 ... Image processing apparatus (noise evaluation apparatus), 21 ... CPU, 22 ... Memory, 23 ... Input device, 24 ... Monitor, 25 ... Control / data Bus, 26 ... component decomposition processing unit, 27 ... SNR calculation unit, 28 ... display processing unit.

Claims (20)

A storage unit for storing medical image data;
An estimation unit that estimates a noise component distribution by approximating a signal component distribution expressed by a polynomial with respect to an observation signal distribution related to a partial region of the medical image;
An image processing apparatus comprising: a noise index calculation unit that calculates a noise index related to a partial area that is the same as or different from the partial area based on the estimated noise component distribution.   The image processing apparatus according to claim 1, wherein the noise index calculation unit calculates a standard deviation of the estimated noise component distribution as the noise index.   The image processing apparatus according to claim 1, wherein the medical image is an MRI image.   The image processing apparatus according to claim 1, wherein the medical image is an MRI image using an array coil.   The image processing apparatus according to claim 1, wherein the medical image is an MRI parallel imaging image using an array coil.   The image processing apparatus according to claim 1, wherein the estimation unit approximates the signal component distribution to the observed signal distribution by a least square approximation method.   The image processing apparatus according to claim 1, wherein the medical image is a two-dimensional or three-dimensional image.   The image processing apparatus according to claim 1, wherein the partial area is automatically set from the medical image by a segmentation method.   The image processing apparatus according to claim 1, wherein the partial area is automatically set on a specific part on the medical image.   The noise index calculation unit calculates, as the noise index, a value obtained by dividing an average value of observation signals related to a partial area at the same position or other position as the partial area by a standard deviation of a noise component distribution of the partial area. The image processing apparatus according to claim 1, wherein:   The noise index calculation unit calculates, as the noise index, a value obtained by dividing a difference between average values of observation signals for two partial areas by a standard deviation of a noise component distribution of the partial areas. The image processing apparatus described.   The image processing apparatus according to claim 1, further comprising a display unit that displays the estimated noise component distribution.   The image processing apparatus according to claim 12, wherein the display unit displays the estimated noise component distribution three-dimensionally.   The image processing apparatus according to claim 12, wherein the display unit displays the estimated noise component distribution after performing luminance conversion.   The image processing apparatus according to claim 12, wherein the display unit converts the estimated noise component distribution into a histogram and displays the histogram.   The image processing apparatus according to claim 12, wherein the display unit simultaneously displays a plurality of noise component distributions having different partial areas.   The image processing apparatus according to claim 12, wherein the display unit displays the calculated noise index together with the noise component distribution.   The image processing apparatus according to claim 1, wherein the noise index calculation unit repeats the approximating process while excluding abnormal values from the noise component distribution.   The image processing apparatus according to claim 18, wherein the noise index calculation unit extracts the abnormal value using a predetermined number times a standard deviation of the noise component distribution as a threshold value. Means for estimating a noise component distribution by approximating a signal component distribution expressed by a polynomial to an observed signal distribution related to a partial region of a medical image;
A program for causing a computer to realize means for calculating a noise index related to a partial area at the same position as that of the partial area or based on the estimated noise component distribution.

Images