Back to the list
Congress: ECR25
Poster Number: C-10229
Type: Poster: EPOS Radiologist (scientific)
Authorblock: G. Tremamunno1, A. Varga-Szemes2, D. Kravchenko2, M. T. Hagar2, U. J. Schoepf2, A. Laghi1, M. Vecsey-Nagy2, T. S. Emrich2; 1Rome/IT, 2Charleston, SC/US
Disclosures:
Giuseppe Tremamunno: Nothing to disclose
Akos Varga-Szemes: Research/Grant Support: Siemens Healtineers
Dmitrij Kravchenko: Nothing to disclose
Muhammad Taha Hagar: Nothing to disclose
Uwe Joseph Schoepf: Research/Grant Support: Siemens Healthineers
Andrea Laghi: Nothing to disclose
Milán Vecsey-Nagy: Nothing to disclose
Tilman Stephan Emrich: Research/Grant Support: Siemens Healthineers
Keywords: Artificial Intelligence, CAD, CT-Angiography, CAD, Arteriosclerosis
Results

Patient characteristics

Out of 97 patients who consented to participate, 32 (26 male [81.3%], 68.6 ± 6.8 years of age) patients were available for final analysis as detailed in Figure 1. The median interval between scans was 14.5 days (range: 7.8-18.2). Details of patient characteristics are summarized in Table 2.

Automatic coronary analysis

A total of 401 segments and 185 plaques were identified and automatically evaluated by the DL algorithm in both the UHR PCD-CT and EID-CT datasets. Among these, 71 (38.4%) were calcified, 113 (61.0%) were partially calcified, and one (0.6%) was non-calcified. In the intra-individual inter-scanner comparison, the ACA showed a significantly lower median overall stenosis value in the UHR PCD-CT datasets compared to the EID-CT datasets (41.0 interquartile range [31.0-54.0] vs. 49.0 [35.0-60.0]; p < 0.001. Mean bias 5.6; LoA, -19.3/30.7). A similar trend was observed in the different composition subgroup analysis, with lower median values for calcified plaques (UHR: 42.0 [32.5-55.0], EID-CT: 51.0 [41.0-62.5]; p < 0.001. Mean bias 8.5; LoA, -17.6/34.5) and partially calcified plaques (UHR: 41.0 [30.0-54.0]; EID-CT: 47.0 [30.0-58.0]; p < 0.001. Mean bias 3.8; LoA, -20.0/27.6).

Comparison with manual coronary assessment

The inter-rater agreement between the two experienced readers was substantial for the EID-CT datasets (κ = 0.75) and almost perfect for the UHR PCD-CT datasets (κ = 0.86). Compared to the two readers' assessment, the ACA demonstrated high accuracy in detecting obstructive stenosis (R1: 90.3% and 95.0% for EID-CT and UHR PCD-CT, respectively; R2: 91.3% and 93.7% for EID-CT and UHR PCD-CT, respectively). The agreement between the ACA and MCA was substantial, with Cohen’s Kappa values of κ = 0.67 and 0.74 for EID-CT and UHR PCD-CT, respectively, for Reader 1, and κ = 0.72 and 0.70 for EID-CT and UHR PCD-CT, respectively, for Reader 2.

Accuracy of stenosis assessment

Compared to QCA, ACA grading based on UHR PCD-CT was more accurate than with EID-CT (mean bias: 8.3 vs. 18.3), with narrower limits of agreement (-41.7 to 25.0 vs. -54.6 to 17.9, respectively).

Conventional coronary assessments took a mean time of 7:17±1:50 minutes to complete compared to 2:12±0:26 minutes for the ACA (p < 0.001). The combined aACA significantly reduced reading times on average by 54% from just MCA (3:09±0:46; p < 0.001).
GALLERY