Introduction

Patient motion during MRI exams is a significant clinical problem, rendering scans sometimes clinically unusable and often requiring rescans¹. There are also indications that even subtle motion artifacts affect structural measurements^2,3. Over the years multiple methods have been proposed for detecting and correcting for patient motion during a scan⁴. Most current approaches require additional dedicated hardware or use of navigator sequences. We introduce a novel motion scoring algorithm compatible with standard FSE protocols that correlates well with a human observer score (R² = 0.91). The algorithm uses the k-space data from two elements of the receiver array and produces signals indicative of discrete motions and of continuous motions. The two signals are used to calculate a combined motion score. From the discrete motion signal the timing of the discrete motions can be calculated.

Methods

We acquired 26 T1-FLAIR 2D FSE brain scans of subjects instructeded to move their head in different motion patterns during the scan. The motion patterns included different discrete motions and/or continuous head motion. The number of discrete motions within each scan varied from 0 to 8. The severity of the motion artifacts depends on the number of movements and their timing. The motion artifacts are more severe for movements that take place while the center part of k-space is being scanned. A scan order (Figure 1) was used consisting of multiple echo trains of length 8. For T1-FLAIR each echo train is acquired sequentially across slices before acquisition of the next echo train. The motion signal is calculated by Fourier transforming a pair of complex coil-intensity-corrected images from two of the coils in the receiver array back into k-space, calculating their difference, projecting along the frequency-encode direction as shown in Figure 2. The projected differences are then summed over all slices (Figure 3a), followed by a self normalization step performed by subtracting and dividing the signal by its minimum value (Figure 3b). The motion signal is then calculated by summing over the different echo trains (Figure 3c). A discrete motion signal (Figure 3d) is calculated by removing the baseline per k-space region from the motion signal. 17 series were used as a development set and 9 series for testing. For each series the degree of motion artifact visible in the images was ranked by 3 observers on a scale of 0-10, 10 being the most severe motion corruption. The average observer value defined the user score. The algorithmic motion score was calculated by linearly fitting the areas under the curves shown in Figure 3c-d to the user ranking on the development set. The final score was taken to be the linearly transformed area under the curve in graph 3c if its value is greater than 3 and the transformed area under the curve of graph 3d otherwise.

Results

The calibrated score was evaluated on a test set of nine T1-FLAIR brain series. The subjects were asked to move their head between 0 to 8 times during the scan, with an average of 2 movements, and at different times. The resulting images received a user motion score ranging from 0 to 10 with an average score of 4.6. The R² value for the correlation between the user score and the calculated score was 0.91. The correlation between the user score and the calculated score on the test set and the development set are shown in Figure 4. Figure 5 shows images, taken from the test set, with different levels of motion artifacts and their related user score and calculated score. One of the series, which was listed as motion free in the protocol, exhibited slight motion artifacts, and received a motion score of 1.4 by the algorithm and a user score of 1.

Discussion

This work presents a novel approach for detecting and scoring the severity of motion artifacts in MR images. The method also provides the timing of discrete motions. This opens the door to multiple future strategies for overcoming patient motion and for answering research questions related to the presence and effect of motion artifacts on quantitative and qualitative results of MRI exams. The work needs to be extended to additional coil settings and scan protocols for developing and validating a consistent motion score.

Conclusion

We have developed a novel method for calculating a motion-artifact severity score for 2D FSE images. The method was tested for scoring the motion artifacts caused by rigid body movements, such as a subject’s head movement during a brain scan. The calculated score closely correlates with human visual perception of the artifacts’ severity. The method can be used to detect the presence of both continuous and discrete motions during the scan. The method can be used to time the discrete motions. The calculated motion score integrates the contribution from both discrete and continuous movements. The relative motion score is a parameter free calculation for a given acquisition protocol. A perceptual motion score can be calculated by a linear transformation applied to the relative motion score following a calibration process.

Acknowledgements

No acknowledgement found.