Table 2.
Summary of external test accuracy with internal validation accuracy for each automated deep learning (AutoDL) model.
| AutoDL model | Accuracy, % (95% CI) | Precision, % (95% CI) | Recall, % (95% CI) | F1 score, % (95% CI) | Training time (minutes) | |
| Neuro-T–based model |
|
|
|
|
826 | |
|
|
Internal validation performance (n=290) | 92.4 (89.3-95.5) | Ma: 92.0 (88.1-95.9); SMb: 93.3 (88.4-98.2) | M: 96.8 (94.3-99.3); SM: 84 (76.8-91.2) | M: 94.4 (91.1-97.7); SM: 88.4 (82.1-94.7) |
|
|
|
External test performance (n=290) | 89.3 (85.1-93.5) | M: 89.9 (84.6-95.2); SM: 88.3 (81.3-95.3) | M: 92.8 (88.3-97.3); SM: 84.0 (76.0-92.0) | M: 91.3 (86.4-96.2); SM: 86.1 (78.6-93.6) |
|
| Neuro-T–based anomaly detection model |
|
|
|
|
20 | |
|
|
Internal validation performance (n=1185) | 80.2 (77.9-82.5) | M: 33.3 (26.6-40.0); SM: 86.2 (84.1-88.3) | M: 23.7 (17.7-29.7); SM: 91.0 (89.2-92.8) | M: 27.7 (21.3-34.1); SM: 88.6 (86.6-90.6) |
|
|
|
External test performance (n=206) | 49.5 (42.7-56.3) | M: 61.8 (53.3-70.3); SM: 40.2 (29.5-50.9) | M: 44.0 (35.3-52.7); SM: 58.0 (47.3-68.7) | M: 51.4 (42.6-60.2); SM: 47.5 (36.6-58.4) |
|
| Create ML–based model 1 |
|
|
|
|
76 | |
|
|
Internal validation performance (n=149) | 81.9 (75.7-88.1) | M: 80.6 (72.2-89.0); SM: 83.9 (75.0-92.8) | M: 89.3 (82.7-95.9); SM: 72.3 (61.4-83.2) | M: 84.7 (77.0-92.4); SM: 77.7 (67.6-87.8) |
|
|
|
External test performance (n=206) | 83.5 (78.4-88.6) | M: 82.3 (75.6-89.0); SM: 86.2 (78.7-93.7) | M: 92.8 (88.3-97.3); SM: 69.1 (59.0-79.2) | M: 87.2 (81.3-93.1); SM: 76.7 (67.5-85.9) |
|
| Create ML–based model 2 |
|
|
|
|
13 | |
|
|
Internal validation performance (n=149) | 81.9 (75.7-88.1) | M: 82.0 (73.8-90.2); SM: 81.7 (72.3-91.1) | M: 86.9 (79.7-94.1); SM: 75.4 (64.9-85.9) | M: 84.4 (76.6-92.2); SM: 78.4 (68.4-88.4) |
|
|
|
External test performance (n=206) | 82.0 (76.8-87.2) | M: 79.7 (72.6-86.8); SM: 87.9 (80.8-95.0) | M: 94.4 (90.4-98.4); SM: 63.0 (52.5-73.5) | M: 86.1 (80.0-92.2); SM: 73.4 (63.8-83.0) |
|
| AutoML Vision–based model |
|
|
|
|
25 | |
|
|
Internal validation performance (n=295) | 84.7 (80.6-88.8) | M: 87.0 (82.3-91.7); SM: 80 (72.2-87.8) | M: 90.2 (86.0-94.4); SM: 74.5 (66.0-83.0) | M: 88.6 (84.0-93.1); SM: 77.2 (69.1-85.3) |
|
|
|
External test performance (n=206) | 83.0 (77.9-88.1) | M: 80.0 (73.0-87.0); SM: 91.1 (84.9-97.3) | M: 96.0 (92.6-99.4); SM: 63.0 (52.9-73.1) | M: 87.3 (81.5-93.1); SM: 74.5 (65.0-84.0) |
|
aM: mucosa-confined lesions.
bSM: submucosa-invaded lesions.