Abstract

Fast and accurate anatomical landmark detection can benefit many medicalimage analysis methods. Here, we propose a method to automatically detectanatomical landmarks in medical images. Automatic landmark detection isperformed with a patch-based fully convolutional neural network (FCNN) thatcombines regression and classification. For any given image patch, regressionis used to predict the 3D displacement vector from the image patch to thelandmark. Simultaneously, classification is used to identify patches thatcontain the landmark. Under the assumption that patches close to a landmark candetermine the landmark location more precisely than patches farther from it,only those patches that contain the landmark according to classification areused to determine the landmark location. The landmark location is obtained bycalculating the average landmark location using the computed 3D displacementvectors. The method is evaluated using detection of six clinically relevantlandmarks in coronary CT angiography (CCTA) scans: the right and left ostium,the bifurcation of the left main coronary artery (LM) into the left anteriordescending and the left circumflex artery, and the origin of the right,non-coronary, and left aortic valve commissure. The proposed method achieved anaverage Euclidean distance error of 2.19 mm and 2.88 mm for the right and leftostium respectively, 3.78 mm for the bifurcation of the LM, and 1.82 mm, 2.10mm and 1.89 mm for the origin of the right, non-coronary, and left aortic valvecommissure respectively, demonstrating accurate performance. The proposedcombination of regression and classification can be used to accurately detectlandmarks in CCTA scans.

 

Introduction (beta)

None

 

Conclusion (beta)

None