Keywords:
Retrospective, Inflammation, Diagnostic procedure, CT, Lung, Chest, Cross-sectional study, Performed at one institution
Authors:
S. S. Westphalen1, F. Soares Torres2, T. S. Garcia1; 1Porto Alegre/BR, 2Toronto/CA
DOI:
10.26044/ecr2020/C-14651
Results
A summary of the results is avaiable in table 2. The most prevalent HRCT pattern was “most consistent with non-IPF diagnosis” (54.5% - 63.6%) (figure 2). There was moderate to high agreement in the classification among the four thoracic radiologists (k 0.61). Experts agreement for a binary score “typical UIP/probable UIP” or “indeterminate for UIP/most consistent with non-IPF diagnosis” was high (k 0.79). When the radiology resident was included in the analysis (total of five observers), the agreement in the diagnostic classification was moderate (k 0.59) and the agreement in the binary score was high (k 0.77). The interobserver agreement for the presence of honeycombing was considered very high among experts (k 0.81) and high among all observers (k 0.69) (figure 3). The k value for the predominant imaging pattern was 0.67 for thoracic radiologists and 0.63 for all observers (figure 4). The observers agreed in at least one of the three diagnostic hypotheses in 16 cases (36.4%), and in the most probable diagnosis in 12 cases (27.3%). There was low agreement in the confidence of the most likely diagnosis (k 0.21).