Skip to main content
. 2021 Oct;11(10):4245–4257. doi: 10.21037/qims-21-142

Table 3. Measurement and classification accuracy by location.

Dataset A (399 cases/3,192 locations) All locations Root Ascending aorta Arch Descending aorta
Cases with initially correct estimation (within 5 mm interval for all measurements) 254/399 (63.7%) NA§
Reviewed cases 145/399 (36.3%)
Reviewed locations 1,160
   Correct measurement 748/1,160 (64.5%) 24/290 (8.3%) 110/145 (75.9%) 361/435 (83.0%) 246/290 (84.8%)
   Wrong measurement 377/1,160 (32.5%) 266/290 (90.3%) 32/145 (22.1%) 69/435 (15.9%) 14/290 (4.8%)
   Missing measurement 35/1,160 (3.0%) 0/290 3/145 (2.0%) 5/435 (1.1%) 30/290 (10.3%)
No change of classification 76/145 (52.4%) 96/145 (66.2%) 134/145 (92.4%) 130/145 (89.7%) NA
Change of classification 69/145 (47.6%) 49/145 (33.8%) 11/145 (7.6%) 15/145 (10.3%) NA
Aneurysms misclassified by DL-algorithm in reviewed cases 34/145 (23.4%) 18/145 (12.4%) 11/145 (7.6%) 11/145 (7.6%) NA
Finally correct measurements (incl. reviewed locations) 2,778/3,192 (87.0%) 537/798 (67.3%) 364/399 (91.2%) 1,123/1,197 (93.8%) 754/798 (94.5%)
Aneurysms misclassified by DL-algorithm (all cases) 34/399 (8.5%) 18/399 (4.5%) 11/399 (2.8%) 11/399 (2.8%) NA

Measurement and classification accuracy by location. , cases with a difference >5 mm between DL and original measurements underwent detailed review of all measurements. , in 6 cases the misclassified aneurysm extended to multiple locations, those cases were only counted once. §, not available in original reports. DL, deep learning; NA, not available.