Method for correcting a diffusion image having an artifact

Abstract

The present invention is related to a method for correcting a diffusion image having an artifact, the method comprising: (a) providing a set of diffusion images comprising the diffusion image having the artifact; (b) calculating a first signal intensity of each image in the set of diffusion images; (c) plotting a graph of serial number of slice of the set of diffusion images versus the first signal intensity; (d) calculating a second signal intensity of the diffusion image having the artifact by performing interpolation on the graph; and (e) correcting the diffusion image having the artifact base on the second signal intensity.

Claims

1. A method for correcting a diffusion image having an artifact, the method comprising: (a) providing a set of diffusion images comprising the diffusion image having the artifact; (b) calculating a first signal intensity of each image in the set of diffusion images; (c) plotting a graph of serial number of slice of the set of diffusion images versus the first signal intensity; (d) calculating a second signal intensity of the diffusion image having the artifact by performing interpolation on the graph; and (e) correcting the diffusion image having the artifact base on the second signal intensity.

2. The method of claim 1, wherein the set of diffusion images are diffusion weighted images, diffusion spectrum images, diffusion tensor images, high angular resolution images, or q ball images.

3. The method of claim 1, wherein the diffusion image having the artifact is caused by subject moving.

4. The method of claim 1, wherein the interpolation is linear interpolation, polynomial interpolation, or spline interpolation.

5. The method of claim 4, wherein the spline interpolation is B-spline interpolation.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) FIG. 1 shows one example of iterative interpolation method. The steps of iterative interpolation are shown in (B) to (D).

(2) FIG. 2 shows reference images (top row) and salvaged images (bottom row) for (A) b=310 s/mm.sup.2 with diffusion direction [0.0601, 0.9895, 0.1314], (B) b=2760 s/mm.sup.2 with diffusion direction [0.3834, 0.5477, 0.7437], and (C) b=4000 s/mm.sup.2 with diffusion direction [0.0419, 0.9022, 0.4292].

(3) FIG. 3 shows representative generalized fractional anisotropy (GFA) maps from a single slice in reference (left), degraded (middle), and corrected (right) DSI data sets.

(4) FIG. 4 shows the plot of number of signal dropout vs. functional difference (FD) in degraded and corrected DSI datasets.

(5) FIG. 5 shows the original images (A), corrected images (B) and the difference between original images and corrected images (C) for two continuous dropouts correction. The top row shows the first dropout and the bottom row shows the second dropout.

SUMMARY OF THE INVENTION

(6) A key problem in diffusion MRI is head motion, which may cause signal attenuation artifacts (signal dropouts) of the image and lead to errors in diffusion index calculation. To prevent these errors, here we provide a correction method which interpolation is used to correct signal dropouts. This method successfully salvaged the dropout images and resumed accurate generalized fractional anisotropy (GFA) calculation.

(7) The present invention provides a method for correcting a diffusion image having an artifact, the method comprising: (a) providing a set of diffusion images comprising the diffusion image having the artifact; (b) calculating a first signal intensity of each image in the set of diffusion images; (c) plotting a graph of serial number of slice of the set of diffusion images versus the first signal intensity; (d) calculating a second signal intensity of the diffusion image having the artifact by performing interpolation on the graph; and (e) correcting the diffusion image having the artifact base on the second signal intensity.

DETAIL DESCRIPTION OF THE INVENTION

(8) Unless otherwise specified, a or an means one or more.

(9) Diffusion MRI is becoming increasingly important for clinical and neuroscience studies owing to its capability to depict microstructural property of the white matter. For more accurate estimation of diffusion index to reflect microstructural property, recent advance in diffusion MRI techniques, such as diffusion spectrum imaging (DSI) and high angular resolution diffusion imaging, acquire diffusion-weighted (DW) images with multiple diffusion sensitivities and directions. Due to the use of strong diffusion gradients, these techniques are sensitive to head motion, and would give rise to signal dropout of the images. This signal dropout may lead to errors in diffusion index calculation (Yendiki, Anastasia; Koldewyn, Kami; Kakunoori, Sita; Kanwisher, Nancy; Fischl, Bruce: Spurious group differences due to head motion in a diffusion MRI study. NeuroImage Volume 88, March 2014, Pages 79-90). Therefore, the present invention aimed to develop a post-processing algorithm to correct artifacts of diffusion images. The correction performance was tested by simulating signal dropout in the in vivo diffusion data.

(10) The present invention provides a method for correcting a diffusion image having an artifact, the method comprising: (a) providing a set of diffusion images comprising the diffusion image having the artifact; (b) calculating a first signal intensity of each image in the set of diffusion images; (c) plotting a graph of serial number of slice of the set of diffusion images versus the first signal intensity; (d) calculating a second signal intensity of the diffusion image having the artifact by performing interpolation on the graph; and (e) correcting the diffusion image having the artifact base on the second signal intensity.

(11) In a preferred embodiment of the present invention, the set of diffusion images are diffusion weighted images, diffusion spectrum images, diffusion tensor images, high angular resolution images, or q ball images.

(12) In a preferred embodiment of the present invention, the diffusion image having the artifact is caused by subject moving.

(13) In a preferred embodiment of the present invention, the interpolation is linear interpolation, polynomial interpolation, or spline interpolation. In a more preferred embodiment of the present invention, the spline interpolation is B-spline interpolation.

EXAMPLES

(14) The examples below are non-limiting and are merely representative of various aspects and features of the present invention.

(15) Imaging

(16) DSI data were acquired on a 3T MRI system (TIM Trio, Siemens, Erlangen) with a 32-channel phased array coil. The DSI pulse sequence used a twice-refocused balanced echo diffusion echo planar imaging sequence (Reese T G, Heid O, Weisskoff R M, Wedeen V J.: Reduction of eddy-current-induced distortion in diffusion MRI using a twice-refocused spin echo. Magn Reson Med. 2003 January; 49(1):177-82), TR/TE=9600/130 ins, FOV=200200 mm.sup.2, matrix size=8080, 56 slices, and 2.5 mm in slice thickness. A total of 102 diffusion encoding gradients with the maximum diffusion sensitivity bmax=4000 s/mm.sup.2 were sampled on the grid points in a half sphere of the 3D q-space with |q|3.6 units (Li-Wei Kuo, Jyh-Horng Chen, Van Jay Wedeen, and Wen-Yih Isaac Tseng. Optimization of diffusion spectrum imaging and q-ball imaging on clinical MRI system. NeuroImage 41 (2008) 7-18).

(17) Signal Dropout Simulation

(18) Fifteen DSI data sets without signal dropout were selected as the reference. These DSI data sets were used to simulate the data sets degraded by signal dropout; the original DW images were replaced randomly with signal dropout images. For each data set, 15 simulated data sets were created using 15 different dropout numbers, i.e. 10, 30, 50, 70, 90, 130, 170, 210, 250, 290, 340, 380, 420, 460 and 500 dropouts. These dropouts were discrete dropouts. Similar method was used for simulating continuous dropouts, e.g. two continuous dropouts.

(19) Dropout Correction

(20) For each volume data corresponding to a specific diffusion encoding, we found that signal intensity along the z (slice) direction was a smooth curve. In the present embodiment, signal dropout at any z location in the curve was resumed by using least-squares fitting of the remaining data with B-spline curves (David Eberly: Least-Squares Fitting of Data with B-Spline Surfaces. Geometric Tools 2005).

(21) For discrete dropouts, the information on either side of the dropout image was maintained, so the interpolation was performed directly to correct the dropout image. For continuous dropouts, the information on either side of the dropout images was incomplete for directly correcting dropout images. Therefore, iterative interpolation method was used to correct continuous dropouts. One example of iterative interpolation method was as follow: When the continuous dropout images were at slice27 and slice28 (FIG. 1A), four steps were used to correct the continuous dropout images:

(22) Step 1: Discard the dropout information of slice27 and slice28, otherwise the dropout information will generate wrong results.

(23) Step 2: Use the information of slice1slice26 and slice29slice56 for interpolating the missing part caused by the dropout of slice27 and slice28 (the interpolated image is seen as the complex of slice27 and slice28, i.e. slice27.5) (FIG. 1B).

(24) Step 3: Use the information of slice1slice26, slice27.5 and slice29slice56 for interpolating the image of slice27 (the information on either side of slice27 is obtained, so the result of the interpolation is better) (FIG. 1C).

(25) Step 4: Use the information of slice1slice27.5 and slice29slice56 for interpolating the image of slice28 and the iterative interpolation method is completed (FIG. 1D).

(26) Validation of Dropout Correction

(27) We used whole brain tract-based automatic analysis (Yu-Jen Chen, Yu-Chun Lo, Yung-Chin Hsu, Chun-Chieh Fan, Tzung-Jeng Hwang, Chih-Min Liu, Yi-Ling Chien, Ming H. Hsieh, Chen-Chung Liu, Hai-Gwo Hwu, Wen-Yih Isaac Tseng: Automatic Whole Brain Tract-Based Analysis Using Predefined Tracts in a Diffusion Spectrum Imaging Template and an Accurate Registration Strategy. Human Brain Mapping 36:3441-3458 (2015)) to obtain generalized fractional anisotropy (GFA) profiles of 76 white matter tract bundles for each DSI dataset. We tested the performance of the correction method by assessing the functional difference (FD) between GFA profiles derived from the reference DSI data set and those derived from the degraded and corrected data sets (Sylvain Gouttard, Casey B. Goodlett Marek Kubicki, and Guido Gerig: Measures for Validation of DTI Tractography. Proc SPIE Int Soc Opt Eng. 2012 February 23; 8314: 83). Functional difference was defined as:

(28) $\mathrm{FD}=\frac{1}{m*n*l}\underset{k=1}{\overset{n}{.Math.}}\underset{j=1}{\overset{m}{.Math.}}\underset{i=1}{\overset{l}{.Math.}}.Math.\frac{f\left(t_i,{\mathrm{tb}}_j,s_k\right)-f_{\mathrm{ref}}\left(t_i,{\mathrm{tb}}_j,s_k\right)}{f_{\mathrm{ref}}\left(t_i,{\mathrm{tb}}_j,s_k\right)}.Math.$

(29) where f(t.sub.i, tb.sub.j, s.sub.k) and f.sub.ref (t.sub.i, tb.sub.j, s.sub.k) were the measured GFA values at step t.sub.i for tract bundle tb.sub.1 in subject s.sub.k derived from the degraded/corrected DSI data sets and reference DSI data sets, respectively, and 1, in, n indicated, the total number of steps of a tract bundle, the total number of tract bundles, and the total number of subjects, respectively.

(30) Results

(31) To prevent errors in GFA calculation due to signal dropout, a conventional way is to discard the data from the analysis. Such an approach would waste many data sets and potentially bias the study population. To address this issue, we provide a method to correct DSI data sets degraded by signal dropout. In the simulated data sets, we showed that this method can successfully salvage the discrete dropout images (FIG. 2). As shown in FIG. 2, we can see that the dropout images were salvaged as images shown in the bottom row which look very similar to the reference images in the top row. As shown in FIG. 3, the degraded GFA map is apparently different from the reference GFA map, whereas the corrected GFA map look very similar to the reference GFA map. As shown in FIG. 4, the degraded FD values increase sharply with the dropout number. By contrast, the corrected FD values increase slightly with the dropout number. The results of GFA maps and FD confirmed that accurate GFA values can be resumed after correction (FIG. 3 and FIG. 4). For continuous dropouts, the method also successfully corrects two continuous dropouts. As shown in FIG. 5, we can see that the corrected images (B) look very similar to the original images (A).