Validation and Feasibility of Ultrafast Cervical Spine MRI Using a Deep Learning-Assisted 3D Iterative Image Enhancement System

Figures & data

Table 1 Acquisition Parameters and Scan Times for Conventional and Ultrafast Protocols

Sequence	Conventional	Ultrafast	Conventional	Ultrafast	Conventional	Ultrafast
	Sagittal T1WI (TSE)		Sagittal T2WI FS (TSE-FS)		Axial T2WI (GRE-FS)
TR/TE (ms)	440/11	440/11	2700/78	2700/78	447/21	383/18
Flip angle (°)	150	150	150	150	30	30
Field of view (mm)	240 x 240	240 x 240	240 x 240	240 x 240	180 x 180	180 x 180
Matrix size	384 x 268	256 x 180	384 x 308	256 x 205	384 x 308	256 x 205
Slice thickness (mm)	3.0	3.0	3.0	3.0	3.0	3.0
Echo train length	3	3	18	18	1	1
Number of averages	4	2	2	1	2	1
Acceleration factor	2	2	2	2	2	2
Scan time (min:s)	2:22	0:47	1:24	0:27	2:28	0:46
Total scan time (min:s)	Conventional: 6:14			Ultrafast: 2:00

Abbreviations: FS, fat suppression; GRE, gradient recalled echo; TE, echo time; TR, repetition time; TSE, turbo spin echo; WI, weighted image.

Figure 1 The example of sagittal T1-weighted (T1WI) and T2-weighted (T2WI), and axial T2-weighted images, obtained from conventional, ultrafast and ultrafast-DL-3DIIE protocols.

Table 2 Qualitative Assessment and Comparison

	Score (1–5)			P-value	Inter-Observer Agreement
	Conventional	Ultrafast-DL-3DIIE	Ultrafast	Conventional vs Ultrafast-DL-3DIIE
Avoidance of Artifacts	2.8 ± 0.6	3.2 ± 0.8	3.0 ± 0.7	0.021*	0.68 (0.62–0.74)
Boundary sharpness	3.8 ± 0.9	3.5 ± 0.7	2.5 ± 0.6	0.12	0.77 (0.73–0.81)
Visibility of lesions	3.8 ± 0.8	3.6 ± 0.7	2.8 ± 0.5	0.44	0.72 (0.69–0.75)
Overall image quality	3.5 ± 0.8	3.4 ± 0.6	2.5 ± 0.5	> 0.90	0.73 (0.68–0.78)

Notes: Cohen’s weighted kappa values (95% confidence interval) for inter-observer agreement are shown. The p values were calculated with two-tailed t-tests, after Bonferroni correction for multiple comparisons. The symbol (*) indicates being significant.

Abbreviation: DL-3DIIE, deep learning-assisted 3D iterative image enhancement.

Table 3 Quantitative Assessment and Comparison

	Measurement			P-value
	Conventional	Ultrafast-DL-3DIIE	Ultrafast	Conventional vs Ultrafast-DL-3DIIE
Sagittal T1-weighted image
SNR (spinal cord)	61.2 ± 21.3	115.4 ± 40.5	58.2 ± 20.1	< 0.001
SNR (CSF)	29.9 ± 9.8	53.4 ± 16.5	32.4 ± 9.3	< 0.001
SNR (vertebrae)	107.3 ± 36.7	200.4 ± 68.7	101.5 ± 39.2	< 0.001
SNR (disc)	61.9 ± 17.3	118.8 ± 39.4	63.2 ± 18.1	< 0.001
Cord/CSF CNR	31.3 ± 8.7	62.0 ± 25.1	25.8 ± 9.4	< 0.001
Sagittal T2-weighted image
SNR (spinal cord)	70.3 ± 25.8	164.6 ± 65.5	73.1 ± 22.4	< 0.001
SNR (CSF)	160.7 ± 56.4	405.4 ± 143.8	152.7 ± 50.5	< 0.001
SNR (vertebrae)	37.8 ± 12.9	91.5 ± 38.7	35.2 ± 11.9	< 0.001
SNR (disc)	41.2 ± 13.4	105.1 ± 39.5	39.1 ± 12.4	< 0.001
Cord/CSF CNR	90.4 ± 31.2	240.8 ± 88.3	89.6 ± 33.2	< 0.001
Axial T2-weighted image
SNR (spinal cord)	84.1 ± 31.8	174.2 ± 61.0	82.2 ± 30.1	< 0.001
SNR (CSF)	125.2 ± 46.9	259.3 ± 78.2	132.4 ± 39.3	< 0.001
SNR (vertebrae)	30.5 ± 11.4	57.1 ± 19.8	29.3 ± 9.2	< 0.001
SNR (disc)	71.8 ± 21.5	156.2 ± 49.5	63.2 ± 18.1	< 0.001
Cord/CSF CNR	41.1 ± 15.7	85.1 ± 28.5	50.2 ± 17.7	< 0.001

Note: The p values were calculated with two-tailed t-tests, after Bonferroni correction for multiple comparisons.

Abbreviations: CNR, contrast noise ratio; CSF, cerebrospinal fluid; DL-3DIIE, deep learning-assisted 3D iterative image enhancement; SNR, signal-to-noise ratio.

Table 4 Integrated Results of Diagnostic Performance by Two Observers

	Count of Cases (With/Without)		P-value Conventional vs Ultrafast-DL-3DIIE	Inter-Protocol Intra-Observer Agreement	Intra-Protocol Inter-Observer Agreement	Inter-Protocol Inter-Observer Agreement
	Conventional	Ultrafast-DL-3DIIE
Central canal stenosis	9/42	10/41	> 0.90	0.85 (0.81–0.89)	0.73 (0.69–0.77)	0.72 (0.68–0.76)
Foraminal stenosis	8/43	9/42	> 0.90	0.87 (0.85–0.89)	0.75 (0.72–0.78)	0.73 (0.70–0.76)
Endplate degeneration	4/47	8/43	0.36	0.67 (0.61–0.73)	0.54 (0.49–0.59)	0.51 (0.47–0.55)
Disc degeneration	47/4	41/10	0.15	0.65 (0.60–0.70)	0.51 (0.44–0.58)	0.50 (0.45–0.55)
Disc herniation	45/6	43/8	0.77	0.88 (0.85–0.91)	0.78 (0.74–0.82)	0.75 (0.72–0.78)
Other pathologies	17/34	14/37	0.67	0.82 (0.78–0.86)	0.68 (0.63–0.73)	0.66 (0.63–0.69)

Notes: Other pathologies included hyperintense cord lesion, retro-odontoid pseudotumor, spinal lipoma, vertebral fracture and vertebral hypointense lesion. Cohen’s weighted kappa values (95% confidence interval) for agreement evaluation are shown. The p values were calculated with chi-squared tests.

Abbreviation: DL-3DIIE, deep learning-assisted 3D iterative image enhancement.

Figure 2 Sagittal T2-weighted images of a participant with central canal stenosis (C3-7 levels, Orange ellipses), were obtained from both conventional (a) and ultrafast-DL-3DIIE (b) protocols. And sagittal T2-weighted images of a participant with foraminal stenosis (C5-6 level, Orange arrows), obtained from both conventional (c) and ultrafast-DL-3DIIE (d) protocols.

Figure 3 Sagittal T2-weighted images of a participant with disc degeneration (C3-4, C4-5 and C5-6 discs, Orange ellipses), were obtained from both conventional (a) and ultrafast-DL-3DIIE (b) protocols. And sagittal T2-weighted images of a participant with disc herniation (C5-6 disc, Orange arrows), obtained from both conventional (c) and ultrafast-DL-3DIIE (d) protocols.

Table 5 Interchangeability Results for Two Protocols

		Inter-Observer Agreement Rate	95% CI of the Individual Equivalence Index (%)
Central canal stenosis	Intra-protocol	52.9%	−2.93 to 1.13
	Inter-protocol	53.9%
Foraminal stenosis	Intra-protocol	60.8%	−1.67 to 3.34
	Inter-protocol	58.8%
Endplate degeneration	Intra-protocol	78.4%	−5.33 to 1.76
	Inter-protocol	79.4%
Disc degeneration	Intra-protocol	92.1%	−2.11 to 5.45
	Inter-protocol	91.1%
Disc herniation	Intra-protocol	76.5%	−2.42 to 4.41
	Inter-protocol	74.5%
Other pathologies	Intra-protocol	82.4%	−3.51 to 4.78
	Inter-protocol	81.4%

Note: Other pathologies included hyperintense cord lesion, retro-odontoid pseudotumor, spinal lipoma, vertebral fracture and vertebral hypointense lesion.

Abbreviation: CI, confidence interval.