Article Figures & Data
Figures
- Fig 1.
Workflow for image rating and usage.
- Fig 2.
CNN architecture used in the experiment. Here NF represents number of filters and FS represents filter size.
- Fig 3.
Representative filter responses from the fourth convolution layer of the CNN (Conv2D_4). Rows 1 and 2, Filter responses for motion-corrupted axial FLAIR/T2* input images, respectively. Rows 3 and 4, Filter responses from axial/sagittal T1 input images without motion, respectively. Filter responses are independent of image contrast and highlight the recognizable motion artifacts in the motion-corrupted images (arrows).
- Fig 4.
Examples of classification performance for 3 series. A few slices are displayed from each series (left), together with the slice ratings for the entire series (right). The numbers at the top left corner of each image represent the slice number.
Tables
Doctor ID No. Series of Insufficient Quality Technician ID T1 (n = 26) T2 (n = 31) T3 (n = 12) T4 (n = 13) Unneeded Rescans Unneeded Recalls Unneeded Rescans Unneeded Recalls Unneeded Rescans Unneeded Recalls Unneeded Rescans Unneeded Recalls D0 24 7 5 7 0 2 14 0 11 D1 13 14 1 19 1 4 5 2 2 D2 35 4 13 4 8 0 23 0 22 D3 24 11 8 13 5 2 13 0 10 D4 28 7 8 6 2 1 16 0 14 D5 30 4 8 4 3 1 19 0 17 Mean ± SD 25.7 ± 7.4 7.8 ± 4 7.2 ± 4 8.8 ± 6 3.2 ± 2.9 1.7 ± 1.4 15 ± 6.1 0.3 ± 0.8 12.7 ± 6.8 Note:—ID indicates identification. The numbers in parenthesis next to the technician identification numbers represent the total numbers of insufficient quality series identified by each rater.
↵a All numbers reported are from the 49 series of the survey. Each series was evaluated twice, assuming that the scan indication was MS and stroke.
Doctor ID No. Series of Insufficient Quality Technician ID T1 (n = 12) T2 (n = 28) T3 (n = 7) T4 (n = 13) Unneeded Rescans Unneeded Recalls Unneeded Rescans Unneeded Recalls Unneeded Rescans Unneeded Recalls Unneeded Rescans Unneeded Recalls D1 2 10 0 26 0 5 0 11 0 D2 13 2 3 15 0 0 6 2 2 D3 8 6 1 22 1 3 3 7 1 D4 11 4 3 18 1 2 6 0 4 D5 7 5 0 21 0 3 3 6 0 Mean ± SD 8.2 ± 4.2 5.4 ± 3 1.4 ± 0.7 20.4 ± 4.2 0.4 ± 0.2 2.6 ± 0.8 3.6 ± 1.1 6 ± 3.4 1 ± 0.4 Note:—ID indicates identification. The numbers in parenthesis next to the technician identification numbers represent the total numbers of insufficient quality series identified by each rater.
↵a All numbers reported are from the 49 series of the survey. Each series was evaluated twice, assuming that the scan indication was multiple sclerosis and stroke.
- Table 3:
Matrix documenting the number of unneeded rescans and recalls created by the DL approach with different thresholds, assuming that series were scanned to rule out MSa
DL (T = 0.1) DL (T = 0.5) DL (T = 0.8) Rescans Recalls Rescans Recalls Rescans Recalls D0 2 8 6 3 9 0 D1 8 3 15 1 21 1 D2 1 18 3 11 4 6 D3 5 10 10 6 14 4 D4 2 11 5 5 7 1 D5 1 13 3 6 6 3 Mean ± SD 3.2 ± 2.8 10.5 ± 2.1 7 ± 4.7 5.3 ± 1.4 10.2 ± 2.6 2.5 ± 0.9 ↵a All numbers are from the 49 test series. Here D0–D5 represent the same individuals as in Tables 1 and 2.
- Table 4:
Matrix documenting the number of unneeded rescans and recalls created by the DL approach with different thresholds, assuming that the series were scanned to rule out strokea
DL (T = 0.5) DL (T = 0.1) DL (T = 5e–4) DL (T = 1e–6) Rescans Recalls Rescans Recalls Rescans Recalls Rescans Recalls D1 25 0 16 0 11 0 4 0 D2 15 1 8 3 3 3 0 7 D3 20 0 13 2 8 2 3 4 D4 17 1 10 3 6 4 2 7 D5 20 1 12 1 7 1 2 3 Mean ± SD 19.4 ± 3.8 0.4 ± 0.2 11.8 ± 1.4 1.8 ± 1.3 7 ± 2.9 2 ± 1.6 2.2 ± 1.5 4.2 ± 3 ↵a All numbers are from the 49 test series. Here D1–D5 represent the same individuals as in Tables 1 and 2. Physician D0, whose ratings were used to train the DL algorithm, is now absent (as in Table 2) because no “stroke” ratings were available for this reader.
