Accuracy of an Artificial Intelligence Deep Learning Algorithm Implementing a Recurrent Neural Network With Long Short-term Memory for the Automated Detection of Calcified Plaques From Coronary Computed Tomography Angiography.

摘要: Purpose The purpose of this study was to evaluate the accuracy a novel fully automated deep learning (DL) algorithm implementing recurrent neural network (RNN) with long short-term memory (LSTM) for detection coronary artery calcium (CAC) from computed tomography angiography (CCTA) data. Materials and methods Under an IRB waiver in HIPAA compliance, total 194 patients who had undergone CCTA were retrospectively included. Two observers independently evaluated image quality recorded presence CAC right (RCA), combination left main anterior descending (LM-LAD), circumflex (LCx) arteries. Noncontrast CACS scans allowed be used cases uncertainty. Heart centerline labeling automatically performed. Presence assessed by RNN-LSTM. algorithm's overall per-vessel sensitivity, specificity, diagnostic calculated. Results absent 84 present 110 patients. As regards CCTA, median subjective quality, signal-to-noise ratio, contrast-to-noise ratio 3.0, 13.0, 11.4. A 565 vessels evaluated. On basis, achieved 93.1% (confidence interval [CI], 84.3%-96.7%), 82.76% (CI, 74.6%-89.4%), 86.7% 76.8%-87.9%), respectively, RCA, 86.4%-97.7%), 95.5% 88.77%-98.75%), 94.2% (CI. 90.2%-94.6%), LM-LAD, 89.9% 80.2%-95.8%), 90.0% 83.2%-94.7%), 85.0%-94.1%), LCx. 92.1% 92.1%-95.2%), 88.9% 84.9%-92.1%), 90.3% 88.0%-90.0%), respectively. When accounting 76.2%, 87.5%, 82.2%, poor-quality data sets 93.3%, 89.2% 90.9%, when rated adequate or higher combined. Conclusion proposed RNN-LSTM demonstrated high CCTA.