Impact of retraining a deep learning algorithm for improving guideline-compliant aortic diameter measurements on non-gated chest CT

• DOI: 10.1016/j.ejrad.2023.111093
• Publication Date: 2023-09-12T06:20:40Z
• Authors: Francesca Lo Piccolo; Daniel Hinck; Martin Segeroth; Jonathan Sperl; Joshy Cyriac; Shan Yang; Saikiran Rapaka; Jens Bremerich; Alexander W. Sauter; Maurice Pradella
• Keywords: Deep Learning; Cross-Sectional Studies; 610; Humans; Tomography, X-Ray Computed; Aorta; Algorithms

Reliable detection of thoracic aortic dilatation (TAD) is mandatory in clinical routine. For ECG-gated CT angiography, automated deep learning (DL) algorithms are established for diameter measurements according to current guidelines. non-ECG gated (contrast enhanced (CE) and non-CE), however, only a few reports available. In these reports, classification as TAD frequently unreliable with variable result quality depending on anatomic location the root presenting worst results. Therefore, this study aimed explore impact re-training previously evaluated DL tool cohort exams.A 995 patients (68 ± 12 years) CE (n = 392) non-CE 603) chest exams was selected which were classified by initial tool. The re-trained version featured improved robustness centerline fitting cross-sectional plane placement. All cases processed version. results radiologist regarding placement measurements. Measurements correctly measured diameters at each whereas false consisted over-/under-estimation diameters.We 8948 exams. performed 8539/8948 (95.5%) correctly. 3765/8948 (42.1%) correct both versions, (best: distal arch 655/995 (66%), worst: Aortic sinus (AS) 221/995 (22%)). contrast, 4456/8948 (49.8%) version, particular (AS: 564/995 (57%), sinotubular junction: 697/995 (70%)). addition, 318 (3.6%) not available previously. A total 228 (2.5%) showed because tilted planes 181 (2.0%) over-/under-segmentations focus AS 137 (14%) n 73 (7%), respectively).Re-training assessment, resulting 95.5% Our data suggests that can be applied even non-ECG-gated including both,