RESUMO
PURPOSE: True automated detection of coronary artery stenoses might be useful whenever expert evaluation is not available, or as a "second reader" to enhance diagnostic confidence. We evaluated the accuracy of a PC-based stenosis detection tool alone and combined with expert interpretation. METHODS: One hundred coronary CT angiography datasets were evaluated with the automated software alone, by manual interpretation (axial images, multiplanar reformations and maximum intensity projections in free double-oblique planes), and by expert interpretation aware of the automated findings. Stenoses ≥ 50 % were noted per-vessel and per-patient, and compared with invasive angiography. RESULTS: Automated post-processing was successful in 90 % of patients (88 % of vessels). When excluding uninterpretable datasets, per-patient sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were 89 %, 79 %, 74 % and 92 % (per-vessel: 82 %, 85 %, 48 % and 96 %). All 100 datasets were evaluable by expert interpretation. Per-patient sensitivity, specificity, PPV and NPV were 95 %, 95 %, 93 % and 97 % (per-vessel: 89 %,98 %, 88 % and 98 %). Knowing the results of automated interpretation did not improve the performance of expert readers. CONCLUSION: Automated off-line post-processing of coronary CT angiography shows adequate sensitivity, but relatively low specificity in coronary stenosis detection. It does not increase accuracy of expert interpretation. Failure of post-processing in 10 % of all patients necessitates additional manual image work-up. KEY POINTS: ⢠Coronary CT angiography is increasingly used for detection of coronary artery stenosis ⢠Computer assisted diagnosis might facilitate and speed up interpretation ⢠Performance in properly segmented cases compared favourably with manual image interpretation ⢠However, automated segmentation failed in about 10 % of cases ⢠Manual reading is still mandatory; computer assisted diagnosis can provide a useful second read.
