Abstract:Fast and accurate anatomical landmark detection can benefit many medical image analysis methods. Here, we propose a method to automatically detect anatomical landmarks in medical images. Automatic landmark detection is performed with a patch-based fully convolutional neural network (FCNN) that combines regression and classification. For any given image patch, regression is used to predict the 3D displacement vector from the image patch to the landmark. Simultaneously, classification is used to identify patches that contain the landmark. Under the assumption that patches close to a landmark can determine the landmark location more precisely than patches farther from it, only those patches that contain the landmark according to classification are used to determine the landmark location. The landmark location is obtained by calculating the average landmark location using the computed 3D displacement vectors. The method is evaluated using detection of six clinically relevant landmarks in coronary CT angiography (CCTA) scans: the right and left ostium, the bifurcation of the left main coronary artery (LM) into the left anterior descending and the left circumflex artery, and the origin of the right, non-coronary, and left aortic valve commissure. The proposed method achieved an average Euclidean distance error of 2.19 mm and 2.88 mm for the right and left ostium respectively, 3.78 mm for the bifurcation of the LM, and 1.82 mm, 2.10 mm and 1.89 mm for the origin of the right, non-coronary, and left aortic valve commissure respectively, demonstrating accurate performance. The proposed combination of regression and classification can be used to accurately detect landmarks in CCTA scans.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: in cardiac CT angiography (CCTA) scans, to achieve fast and accurate automatic detection of anatomical landmarks. Specifically, the authors propose a method based on convolutional neural network (CNN), which automatically detects anatomical landmarks in medical images by combining regression and classification tasks. ### Specific background of the problem 1. **Clinical requirements**: Accurate automatic detection of anatomical landmarks is crucial for many medical image analysis methods, such as those requiring specific anatomical seed points or landmark - based image registration. 2. **Challenges**: Due to anatomical variations among patients and differences in image acquisition methods, accurate automatic detection of anatomical landmarks in medical images is challenging. 3. **Limitations of existing methods**: - Traditional machine - learning methods usually rely on hand - designed features and can only utilize local image information, resulting in time - consuming search schemes to obtain good results. - Deep - learning methods such as convolutional neural network (CNN) can automatically extract features for task description, but most methods only use classification or regression alone for landmark localization. ### The solution proposed in the paper The authors propose a method based on fully convolutional neural network (FCNN), which performs regression and classification tasks simultaneously: - **Regression task**: Predict the 3D displacement vector from the center of the image block to the target landmark. - **Classification task**: Determine whether the image block contains the target landmark. The final landmark position is determined by calculating the average 3D displacement vector of all image blocks classified as containing the landmark. This method assumes that image blocks closer to the landmark can determine the landmark position more accurately than those far from it. ### Experimental verification This method detected six clinically important landmarks in coronary CT angiography (CCTA) scans: - The openings of the right coronary artery and the left coronary artery. - The bifurcation of the left main coronary artery (LM) into the left anterior descending artery (LAD) and the left circumflex artery (LCx). - The origins of the three junctions of the aortic valve. The experimental results show that this method exhibits high accuracy in detecting these landmarks, with the average Euclidean distance errors being: - For the opening of the right coronary artery: 2.19 mm. - For the opening of the left coronary artery: 2.88 mm. - For the LM bifurcation: 3.78 mm. - For the junctions of the right, non - coronary and left aortic valves: 1.82 mm, 2.10 mm and 1.89 mm respectively. ### Conclusion The method proposed in this study achieves high - precision automatic detection of six clinically important anatomical landmarks in CCTA scans by combining regression and classification tasks. This indicates that this method may be suitable for assisting in the detection of coronary artery segments or evaluating the anatomical structure of the aortic root, as a guiding tool for transcatheter aortic valve implantation (TAVI).

CNN-based Landmark Detection in Cardiac CTA Scans

Continuous Extraction of Coronary Artery Centerline from Cardiac CTA Images Using a Regression-Based Method.

Landmark detection in Cardiac Magnetic Resonance Imaging Using A Convolutional Neural Network

Deep Learning-Based Regression and Classification for Automatic Landmark Localization in Medical Images

Coronary Artery Centerline Extraction in Cardiac CT Angiography Using a CNN-Based Orientation Classifier

Landmark Detection in Cardiac MRI by Using a Convolutional Neural Network

A spatio-temporal graph convolutional network for ultrasound echocardiographic landmark detection

Fast and Accurate Craniomaxillofacial Landmark Detection Via 3D Faster R-CNN.

CephaNet: an Improved Faster R-CNN for Cephalometric Landmark Detection.

Cerebrovascular landmark detection under anatomical variations

Automatic Localization of Landmarks in Craniomaxillofacial CBCT Images Using a Local Attention-Based Graph Convolution Network.

Improved Cascade-RCNN for Automatic Detection of Coronary Artery Plaque in Multi-Angle Fusion CPR Images

Fast Multiple Landmark Localisation Using a Patch-based Iterative Network

Automated Anatomical Landmark Detection Ondistal Femur Surface Using Convolutional Neural Network

Anatomic-landmark Detection Using Graphical Context Modelling.

Automatic Detection of Craniomaxillofacial Anatomical Landmarks on CBCT Images Using 3D Mask R-CNN.

A Cascade Regression Model For Anatomical Landmark Detection

Landmark Detection and Coupled Patch Registration for Cardiac Motion Tracking

Automatic Landmark Localization in 3D Medical CT Images: Few-Shot Learning Through Optimized Data Pre-Processing and Network Design.

[Automated cephalometric landmark identification and location based on convolutional neural network]

LitCall: Learning Implicit Topology for CNN-based Aortic Landmark Localization