Deep learning method for aortic root detection

Computed tomography angiography (CTA) is a preferred imaging technique for wide range of vascular diseases. However, extensive manual analysis required to detect and identify several anatomical landmarks clinical application. This study demonstrates the feasibility fully automatic method detecting aortic root, which key landmark in this type procedure. The approach based on use deep learning techniques that attempt mimic expert behavior. A total 69 CTA scans (39 training 30 validation) with different pathology types were selected train network. Furthermore, 71 independently applied as test set assess their performance. accuracy was evaluated by comparing locations marked benchmark (which manually two experts). interobserver error 4.6 ± 2.3 mm. On an average, differences between experts those detected computer 6.6 3.0 mm 6.8 3.3 mm, respectively, when calculated using set. From these results, we can conclude proposed pre-trained CNN models accurately root images without prior segmentation.