JP5416912B2 - Data processing apparatus and medical diagnostic apparatus - Google Patents

Description

Embodiments of the present invention, Ru engages the data processing apparatus and a medical diagnostic apparatus for enhancing SNR (signal to noise ratio) by reducing the noise in the data of the time axis and space axis.

  Conventionally, filtering is performed to reduce random noise existing in data having a space axis or a time axis. In addition to a linear filter whose filter strength does not change temporally and spatially, there is an adaptive filter that determines the filter strength according to data. Structural adaptive filters and SNR adaptive filters have been proposed as adaptive filters that reduce spatial or temporal random noise.

  The structure adaptive filter is a filter that determines the filter strength according to the structure of data and retains the local structure of high-frequency components such as edges, lines, and points. The structure adaptive filter includes a type that detects the direction of an edge or a line and controls the direction of filtering according to the detected direction of the edge or line, or a type that controls the filter strength.

For example, a filter called a sigma filter is known as a structure adaptive filter that controls the filter strength according to an edge detected from image data. The sigma filter creates a weighting function from data in which intermediate frequency components or high frequency components in image data are emphasized, and weights and adds image data and data in which intermediate frequency components or high frequency components are emphasized using the created weight function. This is a filter that performs so-called edge preservation or edge enhancement that reduces noise while preserving edges in image data. Data correction processing (filtering) using this sigma filter is a high-pass filter (HPF: high pass filter) that _converts the original data at the one-dimensional position (x) to be filtered into S _orig (x) and original data S _orig (x). high frequency component (high pass filtered data) obtained by multiplying the filter) by S _high (x), and low frequency component (LPF: low pass filter) obtained by multiplying the original data S _orig (x) If low pass filtered data) is S _low (x), the weighting function is W _high (x), and the correction data after filtering is S _cor (x), Equations (1-1) and (1-2) Can be expressed as

[Equation 1]
W _high (x) = S _high (x) / max [S _high (x)] (1-1)
S _cor (x) = W _high (x) * S _orig (x) + {1-W _high (x)} S _low (x) (1-2)
That is, as shown in Expression (1-1), the _high frequency component S _high (x) is extracted as the edge portion of the original data S _orig (x), and the extracted _high frequency component S _high (x) is the high frequency component S _high Normalized by the maximum value of [x] max [S _high (x)]. The normalized high frequency component is set as a weighting function W _high (x). Next, the correction data S _cor (x) is obtained by weighted addition of the original data S _orig (x) and the low frequency component S _low (x) which is smoothing data using the weight function W _high (x). It is done.

On the other hand, the SNR adaptive filter is a filter that optimizes the filter strength in accordance with the SNR of data. As a specific example of the SNR adaptive filter, Wiener Filter (WF) has been proposed. More specifically, a Fourier WF (FTW) that operates in a normal frequency space and a FREBAS WF (FRW) that operates in a FREBAS space obtained by band division by Fresnel transform have been proposed (see, for example, Non-Patent Document 1). ).
Satoshi Ito, Yoshifumi Yamada: “SNR Improvement Method for MR Video Using Fresnel Transform Dual Method” (English: Ito S, Yamada Y. “Use of Dual Fresnel Transform Pairs to Improve Imaging) ”Med. Imag. Tech. 19 (5), 355-369 (2001)

  However, the conventionally proposed FTW is a filter that improves the SNR of data by processing in a frequency space. In general, the noise component (N) is almost constant in the frequency space, but the signal component decreases as the frequency increases, so there is a problem that if the SNR correction of the data is performed using WF, the deterioration of the high frequency component of the data is inevitable. . On the other hand, since the FREBAS space is a space that maintains some spatial information, the FRW is a filter that can preserve some high-frequency components such as edges compared to the FTW, but it does not work adaptively to the SNR of low-frequency components There is a problem. Thus, an SNR adaptive filter that acts adaptively according to the spatial distribution of SNR over a wide frequency band has not been proposed.

  SNR depends not only on the frequency of the data but also on the location. That is, the SNR is not uniform in the actual data space, and is larger for the high signal portion and smaller for the low signal portion.

  In addition, the SNR may be affected by processing in a display system that visually displays data.

  Furthermore, among various modalities and data subjected to image processing in each modality, there is a data value that does not have a positive correlation with SNR. Examples of data that do not have a positive correlation between the data value and SNR include CT values obtained with X-ray computed tomography (CT) equipment and diffusion coefficients obtained with magnetic resonance imaging (MRI) equipment. And processing data such as (ADC: Apparent Diffusion Coefficient).

  Note that the diffusion weighted signal for obtaining the ADC changes according to the gradient magnetic field factor b and is negatively correlated with the SNR. However, the ADC is calculated by the equation (2) from the signal intensity S (b) of the diffusion weighted signal. For this reason, when the signal strength S (b) of the diffusion weighted signal increases when S (b) <S (0), the ADC value decreases. That is, the SNR of the ADC shows a non-linear correlation with the SNR of the signal strength S (b) of the diffusion weighted signal. Further, the SNR of the ADC has a relationship having a peak when S (0) / S (b) = 3 when viewed from the diffusion weighted signal S (b). Further, when the SNR of the ADC is seen in relation to the value of the ADC, it has a peak when b × ADC = 1.1.

[Equation 2]
ADC = ln {S (0) / S (b)} / b (2)
For this reason, the SNR optimization processing method differs depending on whether the data value and SNR are positively correlated or not. However, at present, no filter has been proposed in which whether or not the data value and the SNR are positively correlated is considered.

The embodiment of the present invention is adaptive to SNR so as to selectively reduce noise while retaining data of a high frequency part and a high SNR part in time axis and space axis data having random noise. An object of the present invention is to provide a data processing apparatus and a medical diagnostic apparatus that can correct the above.

A data processing apparatus according to an embodiment includes a SNR distribution data generation unit that generates SNR distribution data of the processing target data by performing low-pass filtering on the processing target data, and performs a filtering process on the processing target data Filter processing means for generating filter processing data that improves the SNR of the processing target data, weight function creation means for creating a weight function based on information used for window conversion and the SNR distribution data, and the weight function Correction data generating means for generating correction data by performing a weighted calculation of the processing target data and the filter processing data using the .
The data processing apparatus according to another embodiment includes an SNR distribution data generation unit that creates SNR distribution data of the processing target data based on the processing target data, and performs the filtering process on the processing target data. Filter processing means for generating filter processing data that improves the SNR of the target data, weight function creation means for creating a weight function based on information used for window conversion and the SNR distribution data, and using the weight function Correction data creating means for creating correction data by performing a weighted calculation of the processing target data and the filter processing data.

The medical diagnostic apparatus of an embodiment, a data collection means for collecting processed data from the subject, SNR distribution data generating creating SNR distribution data of the processing target data by performing low-pass filtering process on the processing object data Means, filter processing means for generating filter processing data in which an SNR of the processing target data is improved by performing a filtering process on the processing target data, information used for window conversion and the SNR distribution data Weight function creating means for creating a weight function, and correction data creating means for creating correction data by performing a weighted operation on the processing target data and the filter processing data using the weight function .

In the data processing apparatus and the medical diagnosis apparatus according to an embodiment, in the time-axis and space-axis data having random noise, the noise is selectively reduced while retaining the data of the high frequency part and the high SNR part. Data can be corrected adaptively to SNR.

  Embodiments of a data processing apparatus and a medical diagnostic apparatus according to the present invention will be described with reference to the accompanying drawings.

(Configuration and function)
FIG. 1 is a block diagram showing an embodiment of a data processing apparatus according to the present invention.

  The data processing device 1 is constructed by causing a computer 2 to read a program. However, the data processing apparatus 1 can also be configured by providing a circuit having various functions. The data processing apparatus 1 has a function of performing data processing for improving SNR by reducing random noise superimposed on data having at least one of a time axis and a space axis. In particular, the data processing device 1 has a function of adaptively correcting data with respect to SNR so as to selectively reduce noise while retaining data of a high-frequency portion or high SNR portion of data.

  As the processing target data to be corrected by the data processing apparatus 1, any data can be applied as long as it has random noise and has at least one of the time axis and the space axis. For example, the data processing apparatus 1 is built in the medical diagnostic apparatus, and collected data such as raw data, image data, or time axis data collected in the medical diagnostic apparatus can be used as data to be processed by the data processing apparatus 1. However, not only the data obtained in the medical device but also a digital image such as an image taken by a digital camera, a satellite photograph, or a moving image can be used as data to be processed by the data processing apparatus 1.

  Examples of data to be processed with a time axis include electroencephalogram (EEG), electrocardiogram (ECG), electromyogram (EMG), magnetocardiogram (MCG), magnetomyogram (MMG). ) And magnetoencephalogram (MEG). In addition, as an example of processing target data having a spatial axis, data collected by a medical image diagnostic apparatus can be cited. Furthermore, specific examples of the medical image diagnostic apparatus include a simple X-ray diagnostic apparatus, a digital fluorography (DF) apparatus, a computed tomography (CT) apparatus, an MRI apparatus, and a single photon emission computer. Examples include a single photon emission computed tomography (SPECT) apparatus, a positron emission computed tomography (PET) apparatus, and an ultrasonic (US) ultrasonic diagnostic apparatus.

  When data collected by the medical image diagnostic apparatus is set as processing target data, not only image data and time axis data but also projection data can be set as processing target data. The projection data includes projection data obtained in a medical image diagnostic apparatus such as a simple radiography apparatus, a CT apparatus, a SPECT apparatus, a PET apparatus, or an MRI apparatus. Other practical processing target data includes a T1 (longitudinal relaxation time) weighted image (weighted image), a T2 (lateral relaxation time) weighted image obtained by an MRI apparatus, and an ADC.

  Therefore, the data processing apparatus 1 can be incorporated in a medical apparatus such as a medical image diagnostic apparatus or an electroencephalograph, or can be connected to the medical image diagnostic apparatus via a network. FIG. 1 shows an example in which the data processing apparatus 1 is built in the diagnostic imaging apparatus 3.

  The diagnostic imaging apparatus 3 includes a sensor 4, a data storage unit 5, a data processing unit 6, an input device 7 and a display device 8. The sensor 4 has a function of acquiring processing object data by measuring, detecting or receiving the data. When the diagnostic imaging apparatus 3 is an MRI apparatus, an RF (radio frequency) coil is the sensor 4, and when the diagnostic imaging apparatus 3 is an X-ray CT apparatus, the X-ray detector is the sensor 4.

  The data storage unit 5 has a function of storing processing target data acquired by the sensor 4. The data processing unit 6 acquires processing target data from the data storage unit 5 and performs data processing necessary for generating image data in the image diagnostic apparatus 3, and the processing target data after data processing is stored in the data storage device. Has a writing function.

  The data processing apparatus 1 is configured to generate correction data by acquiring processing target data from the data storage unit 5 and performing noise reduction correction, and output the generated correction data to the data storage apparatus. For this purpose, the data processing device 1 includes a data acquisition unit 9, a low-pass filter unit 10, a weight function creation unit 11, an edge enhancement unit 12, and a weighting addition unit 13.

  The data acquisition unit 9 acquires processing target data having random noise spatially or temporally from a medical device such as a medical image diagnostic device or an electroencephalograph (in the example of FIG. 1, a data storage device of the image diagnostic device 3). Thus, the original data has a function to be given to the low-pass filter unit 10, the weight function creation unit 11, and the edge enhancement unit 12. In addition, when the value of the processing target data is nonlinear correlation or negative correlation with the SNR distribution of the processing target data, the data acquisition unit 9 stores the processing target data so that the value of the processing target data is positively correlated with the SNR distribution. A function of converting and giving the original data to the low-pass filter unit 10 and the edge enhancement unit 12 is provided as necessary.

  The low-pass filter unit 10 performs linear or non-linear low-pass filtering on the original data acquired from the data acquisition unit 9, thereby generating low-pass filter processing data with reduced noise, and a low-pass filter processing data as a weight function generation unit. 11 and the weighting addition unit 13.

  The weighting function creating unit 11 obtains SNR distribution data based on the original data obtained from the data obtaining unit 9 and creates a weighting function reflecting the SNR distribution data, and the created weighting function to the weighting adding unit 13. It has the function to give. However, when an instruction to create the SNR distribution and the weight function from the low-pass filter processing data generated in the low-pass filter unit 10 is input from the input device 7 to the data processing device 1, the weight function creation unit 11 The low-pass filter processing data is acquired from the unit 10, and SNR distribution data and a weight function are created based on the low-pass filter processing data.

  The edge enhancement unit 12 acquires the original data from the data acquisition unit 9 when the instruction to perform the edge enhancement processing of the processing target data is input from the input device 7 to the data processing unit 1, and the edge to be stored in the original data , A function to extract an edge part corresponding to a line or a point-like structure part, a function to obtain a weight function for an edge part based on the value of the extracted edge part, a value for the extracted edge part and the obtained edge part A function of giving a weighting function to the weighting addition unit 13; In addition, when an instruction not to perform edge enhancement processing of data to be processed is input from the input device 7 to the data processing device 1, the edge enhancement unit 12 is used for an edge portion that always takes a zero value as necessary. The weighting function is configured to be given to the weighting addition unit 13.

  The weight addition unit 13 uses the weight function acquired from the weight function creation unit 11 to acquire the original data acquired from the data acquisition unit 9 and the low-pass filter processing data acquired from the low-pass filter unit 10 from the edge enhancement unit 12. It has a function of generating correction data in which random noise is reduced by weighting and adding each edge portion of the original data acquired from the edge enhancement unit 12 using a weighting function for the portion. In addition, when the output instruction indicating the output destination is input from the input device 7 to the data processing device 1, the weighted addition unit 13 is configured to output the generated correction data to the designated output destination. In the example of FIG. 1, the weighted addition unit 13 is configured to output correction data to the data storage unit 5 of the image diagnostic apparatus 3. However, the weighted addition unit 13 may be configured to output correction data to a desired device via the display device 8 or a network.

  That is, the data processing apparatus 1 obtains the SNR distribution of the original data from the original data in which the signal strength and the SNR have a positive correlation, and based on the SNR distribution, the higher SNR part has a higher weight and the lower SNR part has a smaller weight. Create a function. Furthermore, low-pass filtering data that has been smoothed by performing low-pass filtering on the original data, and weighted addition of the original data using a weight function according to the SNR distribution, the higher the SNR part, the weaker the intensity. It is possible to obtain correction data that has been low-pass filtered and low-pass filtered with a stronger intensity at a lower SNR portion. In the correction data obtained in this way, the original data is stored in the high SNR portion, and the noise is reduced in the low SNR portion by strong intensity smoothing. That is, the correction data is data obtained by performing non-uniform noise reduction processing on data having non-uniform noise. In addition, edge enhancement can be performed by extracting edge portions from original data and performing weighted addition.

(Operation)
Next, the operation and action of the data processing apparatus 1 will be described.

  Here, a case will be described in which the processing target data is data collected by the image diagnostic apparatus 3 and the weight function is created from the low-pass filter processing data generated by the low-pass filter unit 10.

  First, the processing target data of the subject is collected in advance by the sensor 4 of the image diagnostic apparatus 3, and the collected processing target data is stored in the data storage unit 5. The processing target data stored in the data storage unit 5 is a target of data processing in the data processing unit 6 for generating image data. However, when random noise exists in the processing target data, it is important to perform noise reduction correction on the processing target data during the data processing. Here, depending on which process the noise reduction processing is performed at, the processing target data may have a relationship of SNR distribution and nonlinear correlation (or negative correlation), or a relationship of positive correlation. .

  As a specific example, a case will be described in which projection data collected by an X-ray CT apparatus is used as data to be processed by the data processing apparatus 1.

  FIG. 2 is a flowchart showing the processing procedure of the data processing unit 6 when the diagnostic imaging apparatus 3 shown in FIG. 1 is an X-ray CT apparatus, and the reference numerals with numerals in the figure indicate the steps of the flowchart. .

As shown in FIG. 2, in step S1, the X-ray transmitted through the subject is detected by the X-ray detector which is the sensor 4 of the X-ray CT apparatus. The X-ray detector outputs a transmitted dose distribution I / I ₀ as pure raw data. Next, in step S2, the data processing unit 6 performs preprocessing including logarithmic conversion and sensitivity correction on the transmitted dose distribution I / I ₀ which is pure raw data. Thereby, the transmitted dose distribution I / I ₀ is converted into an integral value of the absorption coefficient μ. In the X-ray CT apparatus, the integrated value of the X-ray absorption coefficient μ after pre-processing is often stored as raw data in the data storage unit 5 instead of the X-ray transmission dose distribution I / I ₀ .

  Next, in step S3, the raw data is subjected to post-processing including water correction in the data processing unit 6 to become water correction data. In step S4, the data processing unit 6 performs back projection processing on the water correction data to obtain back projection data. Next, in step S5, image data for one sheet is generated by image reconstruction processing of a plurality of backprojection data corresponding to one image. Note that pure raw data, raw data, and water correction data before backprojection processing are collectively referred to as projection data.

  FIG. 3 is a diagram showing projection data collected as processing target data of the data processing apparatus 1 when the diagnostic imaging apparatus 3 shown in FIG. 1 is an X-ray CT apparatus.

  Here, for simplicity, a case will be described in which the projection data has a one-dimensional distribution in the x-axis direction perpendicular to the projection direction. Therefore, depending on the data to be processed, there may be a distribution not only in the x-axis direction but also in the y-axis direction and the z-axis direction intersecting with the x-axis. Further, when the processing target data is time axis data, there is a distribution also in the time t-axis direction. The same applies to data shown in each figure such as FIG. 7 having a space axis and a time axis, which will be described later, and is n-dimensional data (n is a natural number) distributed in the x-axis, y-axis, z-axis, and t-axis directions. In some cases.

FIG. 3A is a cross-sectional view of a subject to be detected in the processing target data, and FIG. 3B is a sensor 4 of the X-ray CT apparatus in which X-rays transmitted through the subject shown in FIG. The X-ray transmission dose distribution I / I ₀ at the position x detected by the X-ray detector, FIG. 3C is based on the X-ray transmission dose distribution I / I ₀ shown in FIG. , absorbed dose distribution ln (I ₀ / I) of X-ray at the position x obtained Te FIG. 3 (d) shows a SNR distribution S _snr absorbed dose of X-rays (x) shown in Figure 3 (c).

As shown in FIG. 3A, the cross section of the subject is covered with fat, and bones and organs exist inside. Such an object is irradiated with an X-ray having an incident count value of _I0 per one of a plurality of detection elements included in the X-ray detector. Then, X-rays that have passed through the subject are detected in each X-ray detection element. Then, an X-ray transmission dose distribution I / I ₀ as shown in FIG. 3B is output from the X-ray detector. The X-ray transmission dose distribution I / I ₀ is an incident count value I ₀ per detection element to the subject and a transmission dose that is an X-ray output count value from the subject, that is, one after the subject is transmitted. This is the transmitted dose ratio of the count value I of X-rays received by the detection element.

The X-ray incidence count value I ₀ to the subject and the X-ray output count value I from the subject are expressed by the equation (3) when the X-ray absorption coefficient on a certain projection line (path) p is μ (p). ).
[Equation 3]
I = I ₀ exp [-∫ _p μ (p) dp] (3)

Therefore, the X-ray absorbed dose distribution obtained by logarithmically converting the reciprocal of the X-ray transmitted dose distribution I / I ₀ from Equation (3) is the integral of the absorption coefficient μ (p) as shown in Equation (4). Value.
[Equation 4]
∫ _p μ (p) dp = ln [I ₀ / I] (4)

As shown in FIG. 3B, the X-ray transmission dose distribution I / I ₀ , that is, the signal value of the pure raw data has a positive correlation with the SNR. That is, the X-ray transmission dose distribution I / I ₀ is small in a path that passes through a material having a large absorption coefficient indicating the degree of X-ray absorption, such as a metal material such as bone (calcium) or an artificial bone head. In particular, assuming that the sensitivity of the X-ray detection element is constant between channels, the SNR distribution S _snr of the count value I in each detection element is proportional to the X-ray transmission dose distribution I / I ₀ . That is, Formula (5) is materialized.
[Equation 5]
S _snr ∝I / I ₀ (5)

On the other hand, the absorbed dose distribution ln [I ₀ / I] at the position x indicating the distribution of the X-ray absorption coefficient μ (p), the absorption coefficient μ (x) and the CT value (CT #) after image reconstruction are shown in FIG. As shown in 3 (c) and (d), there is a nonlinear correlation with SNR. Specifically, the absorbed dose distribution ln [I ₀ / I] has a nonlinear relationship with the SNR so that the SNR peaks when I ₀ / I = 3. That is, the SNR is small in a path that passes through a substance having a large absorption coefficient such as bone or a path that hardly absorbs. In the tomographic image of the subject shown in FIG. 3 (a), the CT value distribution is indicated by the luminance.

  Next, as another specific example, a case will be described in which projection data collected by the MRI apparatus is used as data to be processed by the data processing apparatus 1.

  FIG. 4 is a diagram showing projection data collected by a radial scan as data to be processed by the data processing apparatus 1 when the diagnostic imaging apparatus 3 shown in FIG. 1 is an MRI apparatus.

4A is a cross-sectional view of a subject to be detected in the processing target data, and FIG. 4B is a position x collected from the subject shown in FIG. 4A by a radial scan in the MRI apparatus. It is a figure which shows MR (magnetic resonance) signal intensity | strength I (x) = (S) _Sx (p) dp or SNR distribution S _snr (x).

  The radial scan is a scan that collects data in a radial manner passing through the origin in the k space (Fourier space) by changing the gradient magnetic field. In k-space, data orthogonal to the projection direction and passing through the center corresponds to projection data. Therefore, the MR signal collected by the radial scan corresponds to projection data.

As shown in FIG. 4 (a), the cross section of the subject is covered with fat, and bones and organs exist inside. When MR signals are collected from such a subject by a radial scan with a certain direction as a projection direction, the signal intensity I (x) = S _x (p) at a position x perpendicular to the projection direction as shown in FIG. An MR signal having dp or an SNR distribution S _snr (x) is obtained. As shown in FIG. 4A, the signal intensity S _x (p) dp of the MR signal collected by the radial scan has a positive correlation with the normal SNR distribution S _snr (x).

  Further, PROPELLER (periodically rotated overlapping parallel lines with enhanced reconstruction) is known as a data collection method similar to a radial scan. PROPELLER is a technique for collecting data in k-space while rotating a blade, which is a band-like region composed of a plurality of parallel k-space trajectories, around the origin of k-space. The data on the k space collected by PROPELLER does not necessarily pass through the center of the k space, but can be the data to be processed by the data processing apparatus 1 like the data collected by the radial scan. In this case, a set of MR signals having a signal intensity distribution corresponding to the number of parallel data strings existing in the blade is obtained.

  As described above, the absorbed dose distribution μ (p), which is projection data obtained by the projection method in the diagnostic imaging apparatus 3 such as the X-ray CT apparatus, the SPECT apparatus, and the PET apparatus, has a path p having a large value or a small value. The MR data collected by the radial scan corresponding to the projection data in the MRI apparatus has a large magnetization in the signal source, whereas the projection data obtained through the transmission has a nonlinear correlation property that the SNR decreases. Projection data that has passed through a path with higher intensity has a property that SNR improves.

  The data processing apparatus 1 is capable of generating noise regardless of whether the processing target data has a positive correlation with SNR and the processing target data has a nonlinear correlation or negative correlation with SNR. Reduction correction processing can be performed. Therefore, desired data can be given to the data processing apparatus 1 as processing target data. Then, when the processing target data is given to the data processing device 1, it is possible to adaptively adjust the SNR to reduce random noise superimposed on the processing target data.

  Prior to the correction processing of the processing target data, an instruction is given from the input device 7 to the data processing device 1 as to whether or not to perform edge enhancement that preserves and enhances the edge portion of the processing target data. However, whether or not to perform edge enhancement may be determined in advance without depending on the instruction information from the input device 7.

  FIG. 5 is a flowchart showing a processing procedure for performing noise reduction processing adaptively to the SNR with respect to the data value of the processing target data by the data processing device 1 shown in FIG. 1, and FIG. 6 is the data processing shown in FIG. FIG. 4 is a flowchart showing a calculation procedure performed to perform noise reduction processing adaptively to the SNR with respect to the data value of the processing target data in the apparatus 1, and numerals in FIG. Each step is shown. FIG. 7 is a diagram showing an example of low-pass filter processing data, a weighting function, a weighting function for edge portions, and correction data generated by the calculation shown in FIG. 6 in time series.

First, the data acquisition unit 9 acquires predetermined processing target data from the data storage unit 5 of the diagnostic imaging apparatus 3. Here, when the signal strength of the acquired processing target data has a nonlinear correlation or negative correlation with the SNR, the data acquisition unit 9 has a positive correlation between the signal strength of the processing target data and the SNR. The processing target data is converted as follows. Data whose signal intensity has a positive correlation with SNR is _{defined as} original data S _orig (x) at position x for noise reduction correction. This makes it possible to obtain SNR distribution data from the original data S _orig (x) in a later step. In particular, the original data S _orig (x) having a positive correlation can be directly used as an SNR distribution function S _snr (x) described later.

When the signal I (x) and the SNR of the converted data intensity have a negative correlation, the data acquisition unit 9 calculates the reciprocal of the signal I (x), for example, as shown in Equation (6-1). The SNR distribution function S _snr (x) representing the SNR distribution data of the original data S _orig (x) at the position x for noise reduction correction can be used. On the other hand, since the signal intensity I (x) of the projection data at the position x collected by the radial scan in the MRI apparatus as shown in FIG. 4 is positively correlated with the SNR, as shown in the equation (6-2) The SNR distribution function S _snr (x) can be used as it is. Not limited to this example, when the signal strength of the processing target data has a nonlinear correlation with SNR, the processing target data is converted to the original data S _orig (x) or SNR distribution using the function f _snr indicating the nonlinear relationship. The function S _snr (x) can be converted. For example, as shown in FIGS. 3C and 3D, the transmitted dose distribution I / I ₀ corresponding to the integral value of the X-ray absorption coefficient μ collected and stored in the X-ray CT apparatus is processed. Is a logarithmically transformed value ln [I ₀ / I (x)], the signal intensity of the data to be processed has a non-linear correlation with the SNR. Therefore, for example, the data acquisition unit 9 _{converts the} X-ray absorbed dose distribution ln [I ₀ / I (x)] into the SNR distribution function S _snr (x ) Can also be converted.

[Equation 6]
S _snr (x) = 1 / I (x) (6-1)
S _snr (x) = I (x) (6-2)
S _snr (x) = f _snr [ln {I ₀ / I (x)}] (6-3)

On the other hand, when the signal strength of the processing target data has a positive correlation with the SNR, the data acquisition unit 9 directly uses the processing target data as the SNR distribution function S _snr (x) at the position x for noise reduction correction. be able to.

Normally, an X-ray CT system uses the integral value 吸収_p μ (p) dp = ln (I ₀ / I) of the absorption coefficient μ whose signal intensity is in a nonlinear correlation with SNR as the original data S _orig (x). Since it is stored, it is realistic to use raw data that is an integral value of the absorption coefficient μ as data to be processed from the viewpoint of making it unnecessary to install a new storage device or store data.

  When reconstructing image data using a projection method as in an X-ray CT apparatus, it is effective to use data before backprojection processing such as raw data as data to be processed by the data processing apparatus 1. It is. This is because data with a small SNR is evenly distributed on the projection line in the backprojection process, so it is better to perform noise reduction correction before performing the backprojection process because of the SNR degradation and the presence of metal pins. This is because it is possible to reduce the risk of occurrence of artifacts.

  However, image data can also be used as processing target data, and noise reduction processing can be performed in the data processing apparatus 1 in the same manner as when projection data is used as processing target data.

The data acquisition unit 9 gives the original data S _orig (x) acquired in this way to the low-pass filter unit 10 and the edge enhancement unit 12.

Next, in step S <b _> 11 of FIG. 5, the low-pass filter unit 10 performs linear or non-linear low-pass filtering on the original data S _orig (x) acquired from the data acquisition unit 9. As a result, low-pass filter processing data S _low (x) with reduced noise is created. That is, as shown in step S21 of FIG. 6, the low-pass filter unit 10 calculates the _low- pass filter processing data S _low (x) by multiplying the original data S _orig (x) by the low-pass filter H _low (x). .

In FIG. 7A, the horizontal axis indicates the position x, and the vertical axis indicates the signal intensity (SI) of the data. In FIG. 7A, the solid line indicates an example of _low- pass filter processing data S _low (x), and the dotted line indicates an example of original data S _orig (x). As shown in FIG. 7 (a), the low-pass filtered data S _low (x) smoothed by performing low-pass filtering on the original data S _orig (x) having a local signal strength change and noise. Can be created.

  If the low-pass filter is linear, the processing can be simplified. On the other hand, if the low-pass filter is made non-linear, highly accurate noise reduction processing such as locally increasing the strength of smoothing can be performed. Examples of the low-pass filter include an LSI (linear space invariant) filter, a structure adaptive filter, and a Wiener Filter (WF). The LSI filter is a linear filter that has a uniform kernel (filter strength) and does not change in strength temporally and spatially. The structure adaptive filter is a filter that determines a kernel according to the structure of data. WF is a filter that determines the filter strength so that the SNR is optimal in the processing space.

  The filter strength of a filter other than WF is desirably determined using only the filtered signal S or the absolute SNR as an index. The filter strength can be defined by a reduction rate of standard deviation (SD) of Gaussian noise.

Then, the low-pass filter unit 10 gives the low-pass filter processing data S _low (x) to the weighting function creating unit 11 and the weighting adding unit 13.

Next, in step S <b _> 12 of FIG. 5, the weight function creation unit 11 performs SNR representing SNR distribution data of the original data S _orig (x) based on the _low- pass filter processing data S _low (x) acquired from the low-pass filter unit 10. Create the distribution function S _snr (x). This process can be expressed as shown in step S22 of FIG. That is, the SNR distribution function S _snr (x) can be created by converting the _low- pass filter processing data S _low (x) with the function f _snr (S) for creating the SNR distribution. The function f _snr (S) for creating the SNR distribution can be a non-linear conversion function corresponding to the SNR distribution characteristics of the data. The obtained SNR distribution function S _snr (x) can be directly used as the weight function W _snr (x) reflecting the SNR distribution data.

The low-pass filter processing data S _low (x) can be directly used as the SNR distribution function S _snr (x). However, as described above, the original data S _orig (x) may be used as it is as the SNR distribution function S _snr (x). It is also possible to obtain the SNR distribution function S _snr (x) by performing low-pass filtering of the original data S _orig (x) with an intensity different from that of the low-pass filter used to create the _low- pass filter processing data S _low (x). it can. That is, if the value of the original data S _orig (x) is in the relation of SNR and the positive correlation, the original data S _orig (x), the low pass filtered data S _low (x), the original data S _orig (x) or Both the data obtained by nonlinearly transforming the _low- pass filtered data S _low (x) and the data reflecting these characteristics can be used as the SNR distribution function S _snr (x) indicating the SNR distribution.

However, the maximum value of the weight of the weight function W _snr (x) can be set to 1, for example, by normalization. Therefore, the weight function W _{snr is obtained} by normalizing the SNR distribution function S _snr (x) with the maximum value max {S _snr (x)} of the SNR distribution function S _snr (x) by the calculation shown in step S23 of FIG. (x).

Thus, when the weighting function W _snr (x) reflecting the characteristics of the SNR distribution function S _snr (x) is created, the weighting function W has a larger value (weight) in the high SNR part and a smaller value in the low SNR part. _snr (x) can be created. When you create a weight with a normalization function W _snr (x), the maximum value is 1 for the weighting function W _snr (x). Therefore, it is possible to create the weighting function W _snr (x) while reducing the influence of signal intensity variation between different data to be processed. Furthermore, by _creating the SNR distribution function S _snr (x) and the weight function W _snr (x) from the low-pass filter processing data S _low (x), the SNR distribution function W (x) and the weight function W _snr (x) Noise can be reduced.

In FIG. 7B, the horizontal axis indicates the position x, and the vertical axis indicates the weight W. Also, the broken line in FIG. 7B is an example of the weight function W _snr (x), the alternate _{long and short} dash line is normalized to the weight function 1-W _snr (x), and the dotted line is normalized to the original data S _orig (x). An example of data is shown. As shown in FIG. 7 (b), the weight function W _snr (x) created by normalizing the _low- pass filter processing data S _low (x) has a maximum value of 1 and a larger value in the high SNR portion, and the low SNR portion. The smaller the function is.

Furthermore, the SNR distribution function S _snr (x) is not directly used as the weight function W _snr (x), but the SNR distribution function S _snr (x) is nonlinearly transformed by the nonlinear function g as shown in Equation (7). The function W _snr (x) can also be created. If the SNR distribution function S _snr (x) is nonlinearly transformed, the weight of a specific SNR portion can be adjusted.
[Equation 7]
W _snr (x) = g {S _snr (x)} (7)

The nonlinear function g is, for example, a portion where the SNR of the SNR distribution function S _snr (x) is extremely small, that is, when the signal strength S is less than or equal to the threshold Smin, the weight of the weight function W _snr (x) becomes zero, and the signal strength S Is larger than the threshold value Smin, the weight function W _snr (x) can be made such that the weight is smaller as the SNR is smaller and the weight is larger as the SNR is larger. If the weighting function W _snr (x) is created in this way, the original data S _orig (x) whose signal strength S is less than or equal to the threshold value Smin is not stored as it is due to the weighted addition in the _subsequent step, and the low-pass filter processing data S _low Since (x) is satisfied, the smoothing strength for a portion having an extremely small SNR can be increased.

In addition, the non-linear function g is a weight function W that gradually becomes relatively smaller as the weight of the range between the maximum SNR portion and the minimum SNR portion considered as the edge portion becomes farther from the edge portion in a range that is equal to or greater than the threshold value Smin. It can be a function that creates _snr (x). If the weighting function W _snr (x) is created in this way, the proportion of the original data S _orig (x) decreases as the distance from the edge portion decreases due to the weighted addition in the subsequent step, while the low-pass filter processing data S _low Since the ratio of (x) increases, smoothing with stronger intensity is performed in the part far from the edge part. As a result, edge enhancement adapted to the SNR distribution can be performed separately from edge enhancement accompanied with extraction of edge portions.

In the case of the example described above, the nonlinear function g can be determined as shown in Equation (8).
[Equation 8]
g (S) = (S-Smin) ⁿ / Smax: S> Smin, 0: otherwise (8)
However, Smax is the maximum value of the signal strength S, and n is an arbitrary coefficient such that (S) is a downward convex function. Therefore, as n increases, (S) becomes a downward convex function.

  FIG. 8 is a diagram illustrating an example of a nonlinear function g used in the case of creating a weight function by nonlinearly transforming the SNR distribution function in the data processing apparatus 1 shown in FIG.

In FIG. 8A, the vertical axis represents the position x, and the horizontal axis represents the signal strength S. In FIG. 8A, the solid line indicates the SNR distribution function S _snr (x), and the dotted line indicates the original data S _orig (x) (or the normalized original data S _orig (x)). In FIG. 8B, the vertical axis represents the weight W _{snr of the} weight function resulting from the nonlinear transformation g, and the horizontal axis represents the signal strength S. In FIG. 8B, the solid line indicates the nonlinear function W _snr = g (S), and the dotted line indicates the linear function W _snr = S. In FIG. 8C, the vertical axis indicates the weight W _{snr of the} weight function, and the horizontal axis indicates the position x. The solid line in Fig. 8 (c) is the weight function W _snr (x) obtained by nonlinear transformation of the SNR distribution function S _snr (x) using the nonlinear function W = g (S), and the dotted line is The original data S _orig (x) (or the normalized original data S _orig (x)) is shown.

As shown in FIG. 8A, the SNR distribution function S _snr (x) may have an extremely low SNR portion or an edge portion at the end of the position x. In such a case, and FIG. 8 (b) the weight W _snr below a certain signal strength value as shown in a zero, the nonlinear signal strength as relatively weighting W _snr at approximately intermediate value is reduced function W _When the SNR distribution function S _snr (x) is nonlinearly transformed using _snr = g (S), as shown in FIG. 8 (c), the weight W _{snr of the} portion where the SNR is extremely low is zero and the signal intensity is intermediate. A weight function W _snr (x) in which the weight W _{snr of} the part corresponding to the edge part other than the part is emphasized can be created.

On the other hand, as described above, the data obtained by nonlinear conversion of the original data S _orig (x), the low-pass filtered data S _low (x), the original data S _orig (x), or the low-pass filtered data S _low (x) _{Can be used} as the SNR distribution function S _snr (x), and the SNR distribution function S _snr (x) can be used as it is as the weight function W _snr (x).

FIG. 9 shows the data processing apparatus 1 shown in FIG. 1, in which the original data S _orig (x) having a non-linear correlation having a peak with respect to the SNR is converted by a non-linear function f _snr (S) for creating an SNR distribution. It is a figure which shows the example in the case of using SNR distribution function S _snr (x) obtained by this as a weighting function W _snr (x) as it is.

Hereinafter, the case of deriving the non-linear function f _snr that converts the SNR distribution function into an X-ray CT will be described.

The SNR of the projection data lin (I ₀ / I) of the absorbed dose distribution, which is the integral value of the absorption coefficient stored as raw data in the X-ray CT apparatus, is I ₀ / I = 3, that is, ln (I ₀ / I) = It has a non-linear correlation with a peak at 1.1. Similarly, the ADC acquired in the MRI apparatus also has a negative correlation with the signal strength S (b) of the DWI when b> 0, assuming that the signal strength of the DWI with respect to the gradient magnetic field factor b value is S (b). There is a non-linear correlation with a peak when S (0) / S (b) = 3. Further, when the SNR of the ADC is seen in relation to the ADC value, a nonlinear correlation having a peak is obtained when b × ADC = 1.1. The same applies to the T2 weighted image data acquired by the MRI apparatus and the reciprocal of the T2 relaxation time.

Thus, the reference signal value S ₀ and the signal value S (x) at the position x each have random noise, and the signal value S (x) is greater than zero and less than or equal to the reference signal value S _0. In this case, the SNR of the data M (x) at the position x defined by Equation (9) is S (x) / S ₀ = 1/3 with respect to the signal strength of the data M (x), that is, M (x ) = 1.1, it is known to have a characteristic having a maximum value.
[Equation 9]
M (x) =-ln {S (x) / S ₀ } (0 <S (x) ≦ S ₀ ) (9)

The characteristics of SNR can be obtained as follows. That is, if R = S / S ₀ in equation (9), the noise SD of data M (x) is σ _M , and the SD of _R is σ _R , the relationship of equation (10-1) is established, so the data The SNR (M) of M (x) can be obtained as shown in Equation (10-2) using the SNR (R) of R.
[Equation 10]
σ _M = σ _R (δM / δR) = σ _R (1 / R) (10-1)
SNR (M) = M / σ _M = -ln (R) / {σ _R (1 / R)}
= -R × ln (R) / σ _R = -ln (R) × (R / σ _R )
= -ln (R) × SNR (R) (0 <R ≦ 1) (10-2)
In Equation (10-2), under the condition of 0 <R ≦ 1, ln (R) monotonously decreases and SNR (R) monotonically increases with increasing R, so SNR (M) is R = 1 / 3 It has a relationship having a peak when (M (x) = ln (1 / R) = 1.1).

Also, in equation (10-2), when SNR (M) is represented by data M with σ _R = 1, equation (11) is obtained. [Equation 11]
SNR (M) =-ln (R) × R = M × exp (-M) (M> 0) (11)
Here, the SNR distribution function f _snr (M) of the integral value M (or the CT value after reconstruction) of the absorption coefficient in the case of X-ray CT normalizes SNR (M) with the maximum value of SNR (M). Is given by equation (12).
[Equation 12]
f _snr (M) = SNR (M) / Max [SNR (M)] (12)
Substituting equation (11) into equation (12), the SNR distribution function f _snr (M) is given by equation (13).

[Equation 13]
f _snr (M) = e × M × exp (-M) (M> 0) (13)

FIG. 9A shows an example of data M (x) subjected to log conversion defined by equation (9), where the horizontal axis indicates the x-axis and the vertical axis indicates the value of the data M (x). In FIG. 9B, the horizontal axis is represented as data M (= -ln [R]) by the equation (10-2), and the vertical axis is represented as the value of the SNR distribution function f _snr (M) represented by the equation (13). FIG. FIG. 9C shows a weight function W _snr (x) obtained by converting the data M (x) as an SNR spatial distribution by the SNR distribution function f _snr (M).

In this way, when the value of the data M (x) subject to noise correction and the SNR are in a non-linear relationship, the nonlinear function f _snr representing the relationship between the value of the data M (x) and the SNR is used. Thus, the data M (x) such as the X-ray absorption coefficient distribution and the ADC distribution can be converted into the SNR distribution function S _snr (x) based on the SNR distribution of the data M (x).

The weight function W _snr (x) created in this way is given from the weight function creation unit 11 to the weighting addition unit 13.

Next, in step S <b> 13 of FIG. 5, the edge enhancement unit 12 determines whether or not an instruction to perform edge enhancement processing of the processing target data is input from the input device 7 to the data processing device 1. When an instruction to perform edge enhancement processing on the processing target data is input to the data processing apparatus 1, the edge enhancement unit 12 acquires the original data S _orig (x) from the data acquisition unit 9, and the original data In S _orig (x), an edge portion corresponding to an edge, line, or point-like structure portion to be stored is extracted.

That is, in step S24 of FIG. 6, the edge enhancement unit 12 determines whether or not edge enhancement is necessary. If it is determined YES, the edge portion is extracted. Specifically, the edge data S _high (x) of the intermediate frequency component or the high frequency component is obtained by multiplying the original data S _orig (x) by the high pass filter H _high (x) by the calculation shown in step S25 of FIG. Extracted.

Then, the edge enhancement unit 12 in step S14 of FIG. 5, the weighting function for the edge portion for emphasizing an edge portion _{S high (x) W high (} x) is determined from the edge part data S _high (x) . That is, the edge function weight function W _high (x) is created from the intermediate frequency component or high frequency component of the original data S _orig (x). Specifically, for example, the absolute value | S _high (x) | of the edge portion data S _high (x) is changed to the absolute value | S _high (of the edge portion data S _high (x) by the calculation shown in step S26 of FIG. Normalization by the maximum value max {| S _high (x) |} of x) reflects the signal strength characteristics of the edge portion S _high (x), and the weight for the edge portion having the maximum value of 1 The function W _high (x) is created.

In FIG. 7C, the horizontal axis indicates the position x, and the vertical axis indicates the weight W. Further, a broken line in FIG. 7C indicates an example of the weight function W _high (x) for the edge portion, and a dotted line indicates an example of data obtained by normalizing the original data S _orig (x). As shown in FIG. 7C, the edge portion weight function W _high (x) is a function having a maximum value of 1 and having a weight W only at the edge portion.

The edge part data S _high (x) and the edge function weighting function W _high (x) obtained in this way are given from the edge enhancement unit 12 to the weighting addition unit 13.

On the other hand, when the edge enhancement unit 12 determines that an instruction not to perform the edge enhancement processing of the processing target data is input to the data processing apparatus 1, the edge portion data S _high (x) extraction processing and the edge portion weights are performed. The function W _high (x) is not created. However, as shown in step S27 of FIG. 6, when necessary for calculation, zero is substituted for the edge portion weight function W _high (x), and the edge portion weight function W _high ( x) is given to the weighted addition unit 13.

Next, in step S14 of FIG. 5, the weighted addition unit 13 generates correction data S _cor (x) in which random noise is reduced by performing the calculation shown in step S28 of FIG. That is, the weight function W _snr (x) acquired from the weight function creation unit 11 is the weight of the original data S _orig (x), and the weight function 1-W _snr (x) is the low-pass filter processing data S acquired from the low-pass filter unit 10. Add weight as _low (x) weight. Further, when edge enhancement is performed, the edge portion data S _high (x) acquired from the edge enhancement unit 12 is weighted and added using the weight function W _high (x) for the edge portion acquired from the edge enhancement unit 12 as a weight. The

As a result, it is possible to obtain the correction data S _cor (x) in which the noise level is reduced by smoothing with stronger intensity as the SNR of the original data S _orig (x) is smaller. Further, the edge portion can be emphasized by weighted addition of the edge portion data S _high (x).

In FIG. 7D, the horizontal axis indicates the position x, and the vertical axis indicates the signal intensity (SI) of the data. In FIG. 7D, the solid line indicates an example of the correction data S _cor (x), and the dotted line indicates an example of the original data S _orig (x). As shown in FIG. 7 (d), it is possible to obtain correction data S _cor (x) that is smoothed with a stronger intensity as the SNR is lower while the edge is emphasized.

Then, the correction data S _cor (x) created in this way is output from the weighted addition unit 13 to the data storage unit 5 of the diagnostic imaging apparatus 3. However, the correction data S _cor (x) can be output to another device. Then, display image data is reconstructed by data processing on the correction data S _cor (x) in the data processing unit 6 of the image diagnostic apparatus 3. For example, when the correction data S _cor (x) is obtained by correcting the projection data collected in the X-ray CT apparatus, the data processing unit 6 performs post-processing, back-projection processing, and correction processing on the correction data S _cor (x). Image data for display can be created through necessary processing such as image reconstruction processing.

_{Heretofore, an} example has been described in which information regarding window conversion performed during display processing when creating the weight function W _snr (x) is not used, but the processing target data of the data processing apparatus 1 is image data. In some cases, a new weighting function W _snr (x) can be created using information used for window conversion in the display system.

  FIG. 10 is a diagram for performing noise reduction processing adaptively to the SNR for the data value of the processing target data by creating a weight function using information used for window conversion by the data processing apparatus 1 shown in FIG. It is a flowchart which shows a process sequence, and the code | symbol which attached | subjected the number to S in the figure shows each step of each flowchart.

In the flowchart shown in FIG. 10, only a point using information used for window conversion for creating a weighting function and a point for obtaining the luminance value S _disp by _performing WINDOW conversion on the correction data S _cor are shown in FIG. And different. Accordingly, in the flowchart shown in FIG. 10, the same steps as those in the flowchart shown in FIG.

As shown in step S31 of FIG. 10, when the processing target data is image data, the weighting function creating unit 11 performs not only the SNR distribution but also the weighting function W _snr (x ) Can be created. Examples of information used for window conversion used to create the weight function W _snr (x) include window setting values such as window level (WL) and window width (WW), and gamma curves. It is done.

In step S32, the image data which is the correction data S _cor after weighted addition using the weighting function W _snr (x) is _{subjected to} window conversion. In many cases, image data is window-converted based on window setting values (WL, WW), and the signal intensity is displayed as a contrast value that is a luminance value of the display device 8. The window conversion may be non-linear conversion using a gamma curve as well as linear conversion.

  FIG. 11 is a diagram illustrating an example in which image data is subjected to linear window conversion in the data processing unit 6 of the diagnostic imaging apparatus 3 illustrated in FIG. 1.

11A, the vertical axis indicates the position x, the horizontal axis indicates the signal intensity S at the position x, and the solid line in FIG. 11A indicates the image data IMAGE (x) at the position x. In FIG. 11B, the vertical axis represents the contrast value (luminance value) C, and the horizontal axis represents the signal intensity S. In FIG. 11B, the solid line indicates the window conversion function W _disp (S), and the broken line indicates the linear function W _disp = aS + b. In FIG. 11C, the vertical axis represents the contrast value S _disp and the horizontal axis represents the position x. Further, the solid line in FIG. 11C shows the luminance distribution Ic (x) of the display image obtained by performing the linear window conversion on the image data IMAGE (x) using the window conversion function W _disp (S).

Image data IMAGE (x) indicated by the signal intensity S as shown in FIG. 11A is window-converted by the window conversion function W _disp (S) shown in FIG. 11B, and shown in FIG. In this way, the brightness distribution Ic (x) of the display image indicated by the contrast value S _disp is converted. For this purpose, the window setting values WL and WW are arbitrarily determined. When the window setting values WL and WW are determined, the window conversion function W _disp (S) is obtained when the signal intensity S = WL, the signal intensity S becomes the intermediate contrast value S _disp (WL), and the signal intensity S ≧ WL + WW / When the signal strength S is 2, the signal strength S becomes the highest contrast value S _disp (WL + WW / 2), and when the signal strength S ≦ WL-WW / 2, the signal strength S becomes the lowest contrast value S _disp C (WL-WW / 2). Each function is created as a function that can be converted.

Therefore, a window conversion function W _disp (S) is created based on the window setting values (WL, WW), and the SNR distribution function S _snr (x) of the image data IMAGE (x) is expressed by the window conversion function W _disp (S). By _{performing the conversion,} it is possible to create a weight function W _snr (x) in consideration of display processing in the image data display system. That is, the noise reduction correction process can be matched with the visual effect by adapting the weight function W _snr (x) to the display process in the display system.

For example, when the signal intensity S of the image data IMAGE (x) is WL (when S = WL), the window conversion function W _disp (S) has the highest SNR of the luminance distribution Ic (x) of the display image. The policy is that the SNR of the luminance distribution Ic (x) of the display image may be reduced as the signal intensity S of the data IMAGE (x) is separated from the WL and the difference | S-WL | between the signal intensity S and WL increases. Can be determined along.

  FIG. 12 is a diagram illustrating an example of creating a weighting function by converting the SNR distribution function using the conversion function based on the window setting value in step S31 of FIG.

In FIG. 12A, the vertical axis indicates the position x, the horizontal axis indicates the signal intensity S at the position x, and the solid line in FIG. 12A indicates the SNR distribution function S _snr (x) of the image data IMAGE (x). Indicates. Further, FIG. 12 (b) in the vertical axis transformation function W _snr weight W _snr of weighting function is a conversion value of the signal strength S according to (S) (S), the horizontal axis represents the signal intensity S. Further, the solid line in FIG. 12B shows the conversion function W _snr (S). In FIG. 12 (c), the vertical axis represents the weight W _snr (x) of the weight function, and the horizontal axis represents the position x. Also, the solid line in FIG. 12C shows the weighting function W _snr (x) obtained by converting the SNR distribution function S _snr (x) using the conversion function W _snr (S).

By _{converting the} SNR distribution function S _snr (x) as shown in FIG. 12 (a) by the conversion function W _snr (S) shown in FIG. 12 (b) determined according to the above-described policy, FIG. The weight function W _snr (x) adapted to the window conversion process as shown in FIG. Note that the conversion function W _snr (S) shown in FIG. _12B has a weight W of zero when the signal strength S = WL of the SNR distribution function S _snr (x), and the signal strength S ≧ WL + WW / 2. An example in which the function is such that the weight W becomes 1 when the signal strength S ≦ WL−WW / 2. In other words, when the window conversion is linear, the conversion function W _snr (S) can be determined as shown in, for example, Expression (14).

[Formula 14]
W _snr (S) = | S-WL | / (WW / 2): WL-WW / 2 <S <WL + WW / 2, 1: otherwise (14)

Then, correction data S _cor (x) of the image data IMAGE (x) is generated by weighted addition using the weight function W _snr (x) created in this way. Furthermore, a display image in which random noise is reduced from the correction data S _cor (x) and the SNR is adjusted in accordance with the visual effect is generated by the window conversion as shown in FIG. 11 and displayed on the display device 8. The

Note that window conditions such as window setting values (WL, WW) and gamma curves can be arbitrarily set by the user by operating the input device 7. Therefore, when the user changes the window condition, the weight function creating unit 11 can dynamically create the weight function W _snr (x) in synchronization with the set window condition. Further, the correction data S _cor (x) and the display image after the window conversion can also be generated and displayed in synchronization with each other by synchronizing the weight function W _snr (x).

However, once the gamma curve is set in the window conditions, it is not changed frequently. In addition, the window setting values (WL, WW) are often almost determined by the normal data type when displaying images having absolute values such as X-ray CT images or MR images with normalized signal intensity. . Therefore, window conditions such as window setting values (WL, WW) and gamma curves are stored in the data processing apparatus 1 as preset values in advance, and the weight function creation unit 11 automatically calculates the weight function W _snr (x) from the preset values. It can also be created. This eliminates the need to frequently change the window conditions, so that the display image can be created with less processing without dynamically performing data correction processing including the creation of the weighting function W _snr (x) multiple times. Can be displayed.

Separately from this, the window condition is automatically set according to the condition determined arbitrarily in advance, and the weight function creation unit 11 creates the weight function W _snr (x) from the automatically set window condition. It can also be. For example, if the window condition setting condition is set in advance such that the largest image value on the histogram of the image values other than the background such as air is set to WL and twice the WL is set to WW, data processing is performed according to the setting condition. The apparatus 1 or the diagnostic imaging apparatus 3 can automatically set window conditions. For this reason, a display image can be created and displayed by a single data correction process including creation of the weighting function W _snr (x) and weighted addition. By automating the window condition setting in this way, a display image can be created and displayed with less processing.

Furthermore, not only the window conditions, but also the degree of nonlinear transformation necessary to create the weight function W _snr (x) described above and the value of the weight function W _high (x) for the edge portion when edge enhancement is performed Parameters for determining various conditions regarding the data correction processing can be manually adjusted by operating the input device 7. In particular, it may be desirable to be able to adjust the degree of nonlinear transformation and the weight of edge components according to user preferences. Therefore, for example, if dynamic data correction processing can be performed in real time by adjusting a dial such as voice tone control, the user can optimize the display image correction accuracy while referring to the image displayed on the display device 8. Can be

  That is, the data processing apparatus 1 as described above obtains an SNR distribution from given processing target data, and performs linear or non-linear filtering on the processing target data and the processing target data using a weight function reflecting the characteristics of the SNR distribution. Correction data is obtained by performing weighted addition with the applied data.

(effect)
For this reason, in the data processing apparatus 1 described above, even if the data to be processed is data in which the SNR changes locally, the SNR can be improved while adaptively reducing noise according to the SNR. That is, not only the noise reduction but also the storage ratio of the high frequency component can be controlled. In addition, the data processing apparatus 1 can store and emphasize local edge components as necessary.

  Further, since the data processing apparatus 1 can perform processing equivalent to non-linear processing with simple linear processing, high-speed processing is possible. Thereby, dynamic processing in real time can be realized.

  Furthermore, the data processing apparatus 1 has an advantage that the degree of freedom of space to which the correction process can be applied is large. For example, the correction process can be performed in various spaces such as a real space, a projection data space, and a frequency space. That is, since a linear filter such as an LSI filter is used for filtering, even if the data to be processed is data processed by the convolution method in real space or data processed by the FT (Fourier transform) method in frequency space It can be applied as a correction processing target.

  In the data processing device 1, not only when the value of the processing target data and the SNR have a positive correlation, but also when the value of the processing target data and the SNR have a non-linear or negative correlation. It is possible to perform correction processing on the processing target data. That is, when the data to be processed is normal data in which the signal value and the SNR are positively correlated, smoothing can be performed with stronger intensity as the signal value is smaller and the SNR is smaller. On the other hand, when the signal value of the processing target data and the SNR are nonlinearly correlated or negatively correlated, the smoothing intensity can be increased as the signal value is larger and the SNR is smaller.

  In the data processing apparatus 1, a gamma curve and window setting values (WL, WW) for determining the luminance value output from the display device 8 as well as the absolute intensity of the signal, which is a data value such as an image value, are used as the weight function. ) Can be optimized in synchronization with Therefore, it is possible to correct the processing target data in accordance with the visual effect.

  Further, the data processing apparatus 1 has an advantage that other data such as the sensitivity distribution of the sensor 4 is unnecessary because the processing target data is used to obtain the SNR distribution.

  In particular, in recent MRI apparatuses, a plurality of surface coils (surface coils) are often provided as coil elements of the RF coil that is the sensor 4. In this case, since each surface coil has a sensitivity distribution, the data collected by the surface coil will have random noise. Therefore, it is important to perform a correction process for reducing noise caused by sensitivity distribution superimposed on data from a plurality of surface coils. As one method for that purpose, there is a method of obtaining the SNR distribution using the sensitivity distribution data of the surface coil or the estimated value of the sensitivity distribution, and correcting the noise caused by the sensitivity distribution of the surface coil according to the SNR distribution. .

  On the other hand, if the data processing device 1 is used, it is not necessary to estimate the sensitivity distribution data of the surface coil or the sensitivity distribution, and the SNR distribution can be obtained from the data itself from the surface coil. That is, in the data processing device 1, the processing target data and the data obtained by filtering the processing target data are used as a weight function reflecting the SNR distribution. For this reason, even if there is no sensitivity distribution data of the sensor 4 that collects the processing target data, a local SNR distribution can be obtained if the spatial SNR distribution or noise distribution of the processing target data is constant.

  When the SNR distribution is obtained using the sensitivity distribution data, the SNR distribution can be obtained in advance prior to the processing of the processing target data according to the sensitivity distribution data unique to the sensor 4. On the other hand, when the SNR distribution is obtained from the processing target data, the SNR distribution is obtained depending on the processing target data. Therefore, it is necessary to obtain the SNR distribution every time the processing target data is corrected. Therefore, the processing time may be longer when the SNR distribution is obtained from the processing target data than when the SNR distribution is obtained using the sensitivity distribution data.

However, if the weighting function W _snr (x) is created by simple linear filtering in the data processing apparatus 1, the simple filter processing is 1 compared with the case where the weighting function W _snr (x) need not be created. It only increases times. Further, the processing time of the weighted addition process itself is negligible. For this reason, the data processing apparatus 1 can perform high-speed processing as described above.

In addition, when the magnetic resonance signal for each surface coil or each reception channel collected by performing parallel imaging (PI) by an MRI apparatus having a plurality of surface coils as sensors is used as data to be processed by the data processing apparatus 1. In this case, the correction data S _cor (x) may be synthesized after the noise correction processing described above is performed on each processing target data.

  PI is an imaging method in which echo data is received using a plurality of surface coils and phase encoding is skipped to reduce the number of phase encodings necessary for image reconstruction. When echo data is collected by PI, unfolding processing, which is post-processing in PI, is performed on the image data corresponding to each surface coil based on the PI conditions to generate expanded image data. The

  In this case, a plurality of pieces of filter processing data are generated by performing a filtering process on a plurality of pieces of processing target data collected using a plurality of surface coils, and a plurality of pieces of filtering data are generated based on a plurality of SNR distribution data. A weight function is created. Then, a plurality of correction data is created by performing a weighted operation on the plurality of processing target data and the plurality of filter processing data using a plurality of weight functions. Further, a plurality of correction data is synthesized. For this reason, when processing target data from a plurality of sensors or reception channels is corrected, the data processing apparatus 1 is provided with a combining unit for combining a plurality of correction data respectively corresponding to the plurality of processing target data.

  Since there is always a spatial sensitivity distribution in the data for each channel collected by the multi-coil, even if data is collected using a coil with a uniform sensitivity distribution and then uniform filtering is performed to reconstruct the image Even if there is no problem, there is always a low signal part that is buried in noise. For this reason, particularly when PI is performed by an MRI apparatus, the SNR can be improved as compared with the conventional sum of square synthesis using a uniform filter.

An example of an image diagnostic apparatus provided with a multi-channel detector is an X-ray CT apparatus provided with a plurality of sets of X-ray detectors. In this case as well, the correction data S _cor (x) for each channel may be synthesized similarly.

The block diagram which shows embodiment of the data processor which concerns on this invention. The flowchart which shows the process sequence of the data processing part in case the diagnostic imaging apparatus shown in FIG. 1 is an X-ray CT apparatus. The figure which shows the projection data collected as the process target data of a data processor, when the image diagnostic apparatus shown in FIG. 1 is an X-ray CT apparatus. The figure which shows the projection data collected by radial scan as processing target data of a data processor, when the image diagnostic apparatus shown in FIG. 1 is an MRI apparatus. The flowchart which shows the process sequence for performing the noise reduction process adaptively to SNR with respect to the data value of process target data by the data processor shown in FIG. The flowchart which shows the procedure of the calculation performed in order to perform a noise reduction process adaptively to SNR with respect to the data value of process target data in the data processing apparatus shown in FIG. The figure which shows an example of the low-pass filter process data respectively produced | generated by the calculation shown in FIG. 6, a weight function, the weight function for edge parts, and correction data in time series. The figure which shows the example of the nonlinear function used when creating a weight function by nonlinearly transforming an SNR distribution function in the data processor shown in FIG. In the data processing apparatus shown in FIG. 1, the SNR distribution function obtained by converting the original data having a non-linear correlation having a peak relative to the SNR with the non-linear function for creating the SNR distribution is used as it is as the weight function. The figure which shows an example. The flowchart which shows the process sequence for performing a noise reduction process adaptively to SNR with respect to the data value of process target data by producing a weighting function using the information used for window conversion by the data processor shown in FIG. . The figure which shows the example in the case of performing linear window conversion of image data in the data processing part of the diagnostic imaging apparatus shown in FIG. The figure which shows the example in the case of producing a weighting function by converting a SNR distribution function using the conversion function based on a window setting value in step S31 of FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Data processing apparatus 2 Computer 3 Image diagnostic apparatus 4 Sensor 5 Data storage part 6 Data processing part 7 Input device 8 Display apparatus 9 Data acquisition part 10 Low pass filter part 11 Weight function preparation part 12 Edge emphasis part 13 Weighting addition part

Claims (21)

SNR distribution data generating means for creating SNR distribution data of the processing target data by performing low-pass filter processing on the processing target data;
Filter processing means for generating filter processing data in which an SNR of the processing target data is improved by performing filtering processing on the processing target data;
A weight function creating means for creating a weight function based on information used for window conversion and the SNR distribution data;
Correction data creating means for creating correction data by performing a weighted calculation of the processing target data and the filter processing data using the weight function;
A data processing apparatus comprising: The weighting function generating unit, a data processing apparatus according to claim 1, wherein that you create the weighting function by performing a linear transformation to the SNR distribution data. The weighting function generating unit, a data processing apparatus according to claim 1, wherein the benzalkonium create the weighting function by performing a nonlinear transformation to the SNR distribution data. The SNR distribution data generating unit, the data processing apparatus according to claim 1, wherein to Rukoto and the processed data to the projection data having a noise. The SNR distribution data generating unit, the data processing apparatus according to claim 1, wherein said processing target data and to Turkey the image data having a noise. The SNR distribution data generating unit, a computer tomography apparatus, a magnetic resonance imaging apparatus, a positron emission computed tomography apparatus and a single photon said processing the projection data obtained by either emission computed tomography apparatus target data and to Turkey The data processing apparatus according to claim 1. The correction data generation means, the data processing apparatus according to claim 1, wherein that you create the correction data with the enhancement correction of edge in the processed data. The weight function creating means is configured so that the weight is maximum when the signal strength of the SNR distribution data is window level ± window width / 2, and the weight is minimum when the signal strength corresponds to the window level. the data processing apparatus according to claim 1, wherein that you create a weighting function. At least one of window width, window level, and gamma curve is used as information used for the window conversion, synchronized based on the operation of the input device, as a preset value, or automatically according to a predetermined condition Setting means for setting;
Display means for dynamically displaying data generated based on the correction data in synchronization with the operation of the input device when information used for the window conversion is set in synchronization with the operation of the input device When,
The data processing apparatus according to claim 1 , further comprising: The SNR distribution data generating means uses the data collected by using a surface coil having a plurality of sensitivity distributions provided in the magnetic resonance imaging apparatus as the processing target data, and creates the SNR distribution data based on the processing target data. The data processing apparatus according to claim 1. Conversion that converts the processing target data so that the signal strength has a positive correlation with the SNR when the signal strength of the processing target data has a nonlinear correlation or a negative correlation with the SNR of the processing target data the data processing apparatus according to claim 1, further comprising means. The data processing apparatus according to claim 1 , wherein the filter processing means performs the filter processing using an LSI filter, a structure adaptive filter, or a Wiener Filter . The data according to claim 1 , wherein the filter processing means performs the filter processing using a filter having a filter strength determined using only the processing target data or an absolute SNR of the processing target data as an index. Processing equipment. The SNR distribution data generation means creates a plurality of SNR distribution data based on a plurality of processing target data acquired by a plurality of sensors,
The filter processing means generates a plurality of filter processing data by applying a low-pass filter processing to each of the plurality of processing target data,
The weight function creating means creates a plurality of weight functions respectively based on the plurality of SNR distribution data,
The correction data creating means creates a plurality of correction data by performing a weighted operation between the plurality of processing target data and the plurality of filter processing data using the plurality of weight functions,
The data processing apparatus according to claim 1, wherein Rukoto that Yusuke further combining means for combining said plurality of correction data. The SNR distribution data generating unit, by performing non-linear transformation the SNR corresponding to the non-linear correlation processed data is correlated nonlinear with a peak, and features that you create the SNR distribution data The data processing apparatus according to claim 1. SNR distribution data generating means for creating SNR distribution data of the processing target data based on the processing target data;
Filter processing means for generating filter processing data in which an SNR of the processing target data is improved by performing filtering processing on the processing target data;
A weight function creating means for creating a weight function based on information used for window conversion and the SNR distribution data;
Correction data creating means for creating correction data by performing a weighted calculation of the processing target data and the filter processing data using the weight function;
The data processing apparatus according to claim Rukoto to have a. Data collection means for collecting data to be processed from the subject;
SNR distribution data generating means for creating SNR distribution data of the processing target data by applying a low-pass filter process to the processing target data;
Filter processing means for generating filter processing data in which an SNR of the processing target data is improved by performing filtering processing on the processing target data;
A weight function creating means for creating a weight function based on information used for window conversion and the SNR distribution data;
Correction data creating means for creating correction data by performing a weighted calculation of the processing target data and the filter processing data using the weight function;
A medical diagnostic apparatus comprising: At least one of window width, window level, and gamma curve is used as information used for the window conversion, synchronized based on the operation of the input device, as a preset value, or automatically according to a predetermined condition Setting means for setting;
Display means for dynamically displaying data generated based on the correction data in synchronization with the operation of the input device when information used for the window conversion is set in synchronization with the operation of the input device When,
The medical diagnostic apparatus according to claim 17, further comprising: It said data acquisition unit is a medical diagnostic apparatus according to claim 1 7 wherein the benzalkonium to collect raw data as the processing target data. It said data acquisition unit is a medical diagnostic apparatus according to claim 1 7 wherein the benzalkonium to collect image data as the processing target data. It said data acquisition unit is a medical diagnostic apparatus according to claim 1 7 wherein the benzalkonium to collect time axis data as the processing target data.

Images