| Characteristic | N = 811 |
|---|---|
| Sex | |
| F | 41 (51%) |
| M | 40 (49%) |
| Age (years) | 64 (50, 70) |
| Number of exams | |
| 1 | 71 (88%) |
| 2 | 10 (12%) |
| 1 n (%); Median (IQR) | |
Accelerated deep-learning reconstructed cine balanced steady-state free precession cine imaging: Potential for workflow improvement
Statistical analysis
Quantitative variables (age, time per slice, average image quality score) were summarized as median (inter-quartile range, IQR) or mean (standard deviation, SD) and compared by the Wilcoxon rank sum test. Categorical variables (sex, reader-specific score) were summarized as counts (percentages) and compared by the chi-square test or Fisher exact test. Linear regression analysis was used to adjust for age and sex in comparing time per slice between routine and sonic imaging. Inter-reader agreement for image quality scores was assessed by the intra-class correlation (ICC) coefficient (Bartko 1966). P values < 0.05 were considered statistically significant. All analyses were performed in R version 4.4.2 (R Foundation for Statistical Computing, Vienna, Austria).
Results
Patient characteristics
Patient cohort characteristics are summarized in Table 1. There are 10 patients who participated in two studies.
Times per slice
Routine vs Sonic (DL)
The time per slice for planning, SAX, and LAX imaging comparing routine and SONIC is summarized in Table 2. Also compared are age and sex distributions.
| Characteristic | Planning | SAX | LAX | ||||||
|---|---|---|---|---|---|---|---|---|---|
| Routine, N = 431 | Sonic, N = 481 | p-value2 | Routine, N = 421 | Sonic, N = 491 | p-value2 | Routine, N = 341 | Sonic, N = 571 | p-value2 | |
| Time per slice (s) | <0.001 | <0.001 | <0.001 | ||||||
| Median (IQR) | 7.8 (6.6, 9.0) | 4.3 (3.8, 5.4) | 7.9 (6.8, 10.1) | 5.6 (5.1, 7.4) | 7.7 (6.8, 9.7) | 5.7 (4.7, 7.0) | |||
| Mean (SD) | 8.0 (1.9) | 4.7 (1.5) | 8.4 (2.1) | 6.1 (2.1) | 8.2 (2.1) | 5.8 (2.0) | |||
| SONIC type | >0.9 | >0.9 | >0.9 | ||||||
| 1RR | 0 (NA%) | 1 (2.1%) | 0 (NA%) | 2 (4.1%) | 0 (NA%) | 1 (1.8%) | |||
| 3RR | 0 (NA%) | 2 (4.2%) | 0 (NA%) | 3 (6.1%) | 0 (NA%) | 8 (14%) | |||
| 6RR | 0 (NA%) | 45 (94%) | 0 (NA%) | 44 (90%) | 0 (NA%) | 48 (84%) | |||
| Age (years) | 66 (54, 72) | 62 (48, 69) | 0.086 | 67 (53, 72) | 62 (48, 69) | 0.077 | 65 (53, 72) | 62 (50, 69) | 0.3 |
| Sex | 0.047 | 0.028 | 0.070 | ||||||
| F | 17 (40%) | 29 (60%) | 16 (38%) | 30 (61%) | 13 (38%) | 33 (58%) | |||
| M | 26 (60%) | 19 (40%) | 26 (62%) | 19 (39%) | 21 (62%) | 24 (42%) | |||
| 1 n (%); Median (IQR) | |||||||||
| 2 Wilcoxon rank sum test; Fisher’s exact test; Pearson’s Chi-squared test | |||||||||
Figure 1 shows the boxplot of time per slice for planning, SAX, and LAX imaging by routine vs sonic. Data points for 1RR and 3RR sonic are indicated on the graph. For patients who received both routine and sonic imaging (0 for planning, 1 for SAX, and 9 for LAX), a gray line connects the two data points.
- Times significantly shorter for sonic vs routine.
- All “cross-over” patients experienced reduction in time going from routine to sonic.
- 1RR/3RR times tend to be shorter than 6RR. Table 3 below shows that they are indeed significantly shorter.
| Characteristic | Planning | SAX | LAX | ||||||
|---|---|---|---|---|---|---|---|---|---|
| 1RR/3RR, N = 3 | 6RR, N = 45 | p-value1 | 1RR/3RR, N = 5 | 6RR, N = 44 | p-value1 | 1RR/3RR, N = 9 | 6RR, N = 48 | p-value1 | |
| Time per slice (s) | 0.020 | <0.001 | <0.001 | ||||||
| Median (IQR) | 2.6 (2.6, 3.2) | 4.6 (4.0, 5.4) | 3.0 (2.9, 3.0) | 6.2 (5.2, 7.8) | 3.0 (2.7, 3.0) | 6.2 (5.0, 7.3) | |||
| Mean (SD) | 3.0 (0.7) | 4.8 (1.5) | 2.8 (0.5) | 6.5 (1.9) | 3.0 (0.8) | 6.4 (1.7) | |||
| 1 Wilcoxon rank sum test | |||||||||
Adjustment for age and sex
Because Table 2 shows sex and age differences between routine and sonic groups, we adjust for these factors in linear regression analysis. Table 4 shows that the time reductions remain substantial and significant.
| Characteristic | Planning | SAX | LAX | ||||||
|---|---|---|---|---|---|---|---|---|---|
| Diff | 95% CI1 | p-value | Diff | 95% CI1 | p-value | Diff | 95% CI1 | p-value | |
| Exam | |||||||||
| Routine | — | — | — | — | — | — | |||
| Sonic | -3.4 | -4.1, -2.6 | <0.001 | -2.3 | -3.2, -1.4 | <0.001 | -2.4 | -3.4, -1.5 | <0.001 |
| Age (years) | -0.01 | -0.03, 0.01 | 0.3 | -0.02 | -0.05, 0.01 | 0.13 | -0.01 | -0.04, 0.02 | 0.4 |
| Sex | |||||||||
| F | — | — | — | — | — | — | |||
| M | 0.09 | -0.65, 0.82 | 0.8 | 0.30 | -0.61, 1.2 | 0.5 | -0.30 | -1.2, 0.58 | 0.5 |
| 1 CI = Confidence Interval | |||||||||
Image quality
Inter-reader agreement
As shown in Table 5, the inter-reader agreement as measured by the intra-class correlation coefficient (ICC) is strong for routine but weaker for sonic.
| Imaging | Exam | ICC (95%) | p-value |
|---|---|---|---|
| LAX | Routine | 0.351 (0.023, 0.612) | 0.018 |
| LAX | Sonic | 0.049 (-0.211, 0.303) | 0.357 |
| SAX | Routine | 0.418 (0.101, 0.659) | 0.006 |
| SAX | Sonic | 0.146 (-0.136, 0.407) | 0.154 |
Routine vs DL
The image quality scores for routine and sonic imaging are summarized in Table 6. The scores are significantly higher for routine than sonic imaging.
| Characteristic | Reader 1 | Reader 2 | ||||
|---|---|---|---|---|---|---|
| Routine1 | Sonic1 | p-value2 | Routine1 | Sonic1 | p-value2 | |
| SAX | <0.001 | <0.001 | ||||
| 1 | 0 (0%) | 0 (0%) | 0 (0%) | 0 (0%) | ||
| 2 | 15 (44%) | 0 (0%) | 4 (9.5%) | 0 (0%) | ||
| 3 | 17 (50%) | 2 (4.1%) | 32 (76%) | 2 (4.1%) | ||
| 4 | 2 (5.9%) | 44 (90%) | 6 (14%) | 21 (43%) | ||
| 5 | 0 (0%) | 3 (6.1%) | 0 (0%) | 26 (53%) | ||
| LAX | <0.001 | <0.001 | ||||
| 1 | 1 (2.9%) | 0 (0%) | 1 (2.9%) | 0 (0%) | ||
| 2 | 20 (59%) | 0 (0%) | 2 (5.9%) | 0 (0%) | ||
| 3 | 11 (32%) | 2 (3.5%) | 24 (71%) | 2 (3.5%) | ||
| 4 | 2 (5.9%) | 49 (86%) | 7 (21%) | 19 (33%) | ||
| 5 | 0 (0%) | 6 (11%) | 0 (0%) | 36 (63%) | ||
| 1 n (%) | ||||||
| 2 Fisher’s exact test | ||||||
Average reader scores are summarized in Table 7. The scores are significantly higher for routine than sonic imaging.
| Characteristic | SAX | LAX | ||||
|---|---|---|---|---|---|---|
| Routine, N = 42 | Sonic, N = 49 | p-value1 | Routine, N = 34 | Sonic, N = 57 | p-value1 | |
| Average score | <0.001 | <0.001 | ||||
| Median (IQR) | 3.0 (2.5, 3.0) | 4.5 (4.0, 4.5) | 2.5 (2.5, 3.0) | 4.5 (4.0, 4.5) | ||
| Mean (SD) | 2.8 (0.5) | 4.3 (0.4) | 2.8 (0.6) | 4.3 (0.4) | ||
| 1 Wilcoxon rank sum test | ||||||
Summary
- Sonic imaging significantly reduces time per slice compared to routine imaging (even after adjusting for patient age and sex).
- Within sonic, 1RR/3RR imaging is faster than 6RR imaging.
- Sonic images have significantly higher quality than routine images by two readers.
References
Bartko, John J. 1966. “The Intraclass Correlation Coefficient as a Measure of Reliability.” Psychological Reports 19 (1): 3–11. https://doi.org/10.2466/pr0.1966.19.1.3.