I. Visual inspection and SNR assessment
Figure 1 illustrates the pre-processed images that were acquired with different acceleration factors. SENSE-2 images appear substantial quality in all three planes (i.e., sagittal, coronal, and axial) – all structures are clearly visible. In contrast, SENSE-4 images are highly corrupted by noise, particularly at brain stem and cerebellum areas. Fact that these two areas are not well-localized by coil sensitivity, resulting in low SNR and high g-factor amplification. CS-4 enables minimizing ill-posed inverse problem – noise level appears much lower than SENSE-4, but only slightly higher than in SENSE-2, whereas its scan time is reduced half. In comparison with SENSE-2, Fig. 2 demonstrates SNR drops in SENSE-4 for all subcortical structures (p < 0.01). As expected, SENSE-2 provides higher SNR values than these values in SENSE-4 and CS-4 for all observed structures. CS-4 was slightly degraded – significant SNR drops were detected only at Pallidum and Amygdala (p < 0.05). Two largest SNR drops in SENSE-4 were detected at Pallidum and Accumbens with values of 47.47% and 51.96%, respectively. Meanwhile, the SNR drops in CS-4 were detected at Pallidum and Amygdala with values of 7.34% and 10.55%, respectively.
II. Volumetric assessments
In Table 1, the results were obtained by Freesurfer. For SENSE-2 versus CS-4, Wilcoxon sign-rank statistic demonstrates that there were no differences of volumes between these two acquisitions for all subcortical structures. These subcortical volumes were in strong correlation indicated by Pearson’s correlation coefficients (r), in which r-values are larger than 0.90. Moreover, their ES scores are mostly in trivial to small range with absolute values of 0.07–0.23. The moderate volume difference was only detected at Thalamus, Pallidum, and Amygdala with absolute values of 0.57–0.75. For SENSE-2 versus SENSE-4, there were four subcortical volumes that were significantly different including Thalamus, Pallidum, Amygdala, and Accumbens (p < 0.05). Although their correlation coefficients appeared strong with r-values of higher than 0.8, the volume differences are somewhat large with absolute scores of higher than 1.5. Only Putamen had strong correlation and trivial volume difference with r-value and ES-score of 0.99 and 0.01, respectively.
Table 1
Mean and SD of seven subcortical volumes in unit of millimeter cube calculated by using Freesurfer package. Wilcoxon sign-rank test was implemented to assess volume differences in SENSE-4 and CS-4 compared to SENSE-2. p < 0.05 indicates statistical significance.
Brain Structures (n = 33) | SENSE-2 (mm.3) | SENSE-4 (mm.3) | CS-4 (mm.3) | SENSE-2 Vs. SENSE-4 | SENSE-2 vs. CS-4 |
Mean ± SD | Mean ± SD | Mean ± SD |
Caudate | 7155.44 ± 803.34 | 7071.3 ± 847.1 | 7142.82 ± 797.54 | p = 0.63 r = 0.97 ES = 0.43 | p = 0.89 r = 0.97 ES = 0.07 |
Putamen | 10420.19 ± 881.5 | 10420.74 ± 825.47 | 10465.58 ± 872.34 | p = 0.97 r = 0.99 ES < 0.01 | p = 0.86 r = 0.97 ES = − 0.23 |
Thalamus | 14271.89 ± 1188.74 | 15828.83 ± 1489.01 | 14452.88 ± 1243.32 | p < 0.05 r = 0.98 es = -3.97 | p = 0.46 r = 0.98 ES = -0.79 |
Pallidum | 3851.66 ± 379.81 | 3571.62 ± 327.06 | 3931.75 ± 391.53 | p < 0.05 r = 0.87 ES = 1.52 | p = 0.31 r = 0.96 ES = − 0.75 |
Hippocampus | 7605.89 ± 625.92 | 7364.55 ± 593.31 | 7600.77 ± 614.12 | p = 0.14 r = 0.97 ES = 1.77 | p = 0.93 r = 0.96 ES = 0.03 |
Amygdala | 3082.72 ± 281.16 | 2643.55 ± 238.18 | 3030.08 ± 273.41 | p < 0.05 r = 0.86 ES = 3.06 | p = 0.43 r = 0.94 ES = 0.57 |
Accumbens | 1183.98 ± 128.62 | 985.99 ± 124.71 | 1174.44 ± 129.78 | p < 0.05 r = 0.84 ES = 2.74 | p = 0.75 r = 0.94 ES = 0.22 |
ES: effect size, r: Pearson’s correlation coefficient, SD: standard deviation, CS: compressed SENSE |
In Table 2, the results were obtained by FSL’s FIRST. There was only Amygdala obtained by SENSE-4 has significant different (p < 0.05) with moderate correlation (r = 0.55) and large volume difference (ES = -1.29). Although the other six volumes obtained by SENSE-4 had no statistical significance, the moderate correlation was found at Accumbens (r = 0.63). Moreover, large volume differences were also detected at Putamen and Thalamus with absolute ES-scores of 1.09 and 0.94, respectively. In contrast, correlation coefficients associated with SENSE-4 versus CS-4 are very high for all subcortical structures (0.73 < r < 0.98). In addition, almost ES-scores were in the trivial range (0.03 < ES < 0.2). Only Amygdala had moderate volume difference with absolute ES-score of 0.66.
Table 2
Mean and SD of seven subcortical volumes in unit of millimeter cube calculated by using FSL’s FIRST package. Wilcoxon sign-rank test was implemented to assess volume differences in SENSE-4 and CS-4 compared to SENSE-2. p < 0.05 indicates statistical significance.
Brain Structures (n = 33) | SENSE-2 (mm.3) | SENSE-4 (mm.3) | CS-4 (mm.3) | SENSE-2 vs. SENSE-4 | SENSE-2 vs. CS-4 |
Mean ± SD | Mean ± SD | Mean ± SD |
Caudate | 7079.7 ± 672.73 | 7136.97 ± 679.13 | 7072.73 ± 699.56 | p = 0.56 r = 0.86 ES = -0.15 | p = 0.92 r = 0.94 ES = 0.03 |
Putamen | 9947.58 ± 710.16 | 9676.9 ± 7731.11 | 9970.38 ± 52.05 | p = 0.12 r = 0.94 ES = 1.09 | p = 0.96 r = 0.95 ES =-0.08 |
Thalamus | 14711.21 ± 1018.17 | 15046.36 ± 1081.03 | 14747.88 ± 1073.6 | p = 0.18 r = 0.94 ES = -0.94 | p = 0.75 r = 0.98 ES =-0.16 |
Pallidum | 3201.52 ± 306.83 | 3265.15 ± 332.04 | 3210.61 ± 269.87 | p = 0.43 r = 0.82 ES = -0.33 | p = 0.91 r = 0.94 ES =-0.08 |
Hippocampus | 7841.21 ± 763.4 | 7563.33 ± 653.57 | 7777.27 ± 738.21 | p = 0.08 r = 0.78 ES = 0.58 | p = 0.73 r = 0.84 ES = 0.15 |
Amygdala | 2132.732 ± 99.58 | 2531.21 ± 348.75 | 2255.76 ± 371.95 | p < 0.05 r = 0.55 ES = -1.29 | p = 0.12 r = 0.86 ES = -0.66 |
Accumbens | 997.27 ± 145.46 | 968.79 ± 167.87 | 974.85 ± 154.66 | p = 0.26 r = 0.63 ES = 0.21 | p = 0.43 r = 0.73 ES = 0.20 |
ES: effect size, r: Pearson’s correlation coefficient, SD: standard deviation, CS: compressed SENSE |
In Table 3, the results were obtained by volBrain. For SENSE-2 versus SENSE-4, there were five among seven subcortical volumes including Putamen, Pallidum, Thalamus, Amygdala, and Accumbens that had significant differences (p < 0.05). Moderate and strong correlations were detected at those structures (0.59 < r < 0.87), albeit very large volume differences – absolute ES-scores were 1.26, 2.53, 3.94, 3.52, and 3.01, respectively. For SENSE-2 versus CS-4, there were no significant differences for all subcortical volumes. Pearson’s correlation coefficients and effect size scores were mostly in very strong and trivial ranges, respectively.
Table 3
Mean and SD of seven subcortical volumes in unit of millimeter cube calculated by using volBrain package. Wilcoxon sign-rank test was implemented to assess volume differences in SENSE-4 and CS-4 compared to SENSE-2. p < 0.05 indicates statistical significance.
Brain Structures (n = 33) | SENSE-2 (mm.3) | SENSE-4 (mm.3) | CS-4 (mm.3) | SENSE-2 vs. SENSE-4 | SENSE-2 vs. CS-4 |
Mean ± SD | Mean ± SD | Mean ± SD |
Caudate | 7420.91 ± 870.16 | 7244.24 ± 800.05 | 7401.52 ± 840.6 | p = 0.41 r = 0.95 ES = 0.66 | p = 0.88 r = 0.99 ES = 0.16 |
Putamen | 8927.88 ± 760.51 | 8417.27 ± 765.6 | 8910.91 ± 752.44 | p < 0.05 r = 0.86 ES = 1.26 | p = 0.91 r = 0.98 ES = 0.12 |
Thalamus | 11687.88 ± 780.19 | 10568.79 ± 893.7 | 11745.45 ± 863.63 | p < 0.05 r = 0.87 ES = 2.53 | p = 0.64 r = 0.97 ES =-0.25 |
Pallidum | 2361.52 ± 234.9 | 1327.88 ± 319.27 | 2247.27 ± 280.54 | p < 0.05 r = 0.59 ES = 3.94 | p = 0.10 r = 0.93 ES = 1.07 |
Hippocampus | 7573.33 ± 680.37 | 7352.12 ± 649.45 | 7610.3 ± 677.65 | p = 0.17 r = 0.94 ES = 0.94 | p = 0.79 r = 0.99 ES =-0.55 |
Amygdala | 1550.91 ± 175.9 | 1147.58 ± 157.14 | 1506.97 ± 183.12 | p < 0.05 r = 0.77 ES = 3.52 | p = 0.29 r = 0.90 ES = 0.54 |
Accumbens | 652.73 ± 80.28 | 478.79 ± 82.94 | 661.52 ± 85.55 | p < 0.05 r = 0.71 ES = 3.01 | p = 0.57 r = 0.95 ES =-0.32 |
ES: effect size, r: Pearson’s correlation coefficient, SD: standard deviation, CS: compressed SENSE |
In addition, one processed data set was randomly picked up for 3D visualization. The Amygdala obtained from SENSE-4 images was found in irregular shape compared to SENSE-2 and CS-4 as highlighted by white circle in Fig. 3. This evidence can support the consistent results in Table 1–3.
III. Bias and limit of agreement between acquisition techniques
Figure 4 and Fig. 5 show Bland-Altman plots for seven subcortical volumes obtained from SENSE-2 versus CS-4 and SENSE-2 versus SENSE-4, respectively. Generally, there should be three plots for each subcortical structure according three post-processing packages. However, one of them that has smallest bias was shown here. For SENSE-2 versus CS-4 (Fig. 4), minimum bias was found at Hippocampus - mean of difference was only − 5.1 mm3 and limit of agreement (mean ± 1.96 SD) was [-310 300] mm3. Mean differences and limits of agreements for the other subcortical volumes were less than 44 mm3 and [-400 470] mm3, respectively. In contrast, for SENSE-2 versus SENSE-4 (Fig. 5), minimum bias was found at Accumbens - mean of difference was − 28.0 mm3 and limit of agreement (mean ± 1.96 SD) was [-290 240] mm3. Mean differences and limits of agreements for the other subcortical volumes were up to -400 mm3 and [-360 1000] mm3, respectively.