Abstract:This work reports the empirical performance of an automated medical landmark detection method for predict clinical markers in hip radiograph images. Notably, the detection method was trained using a label-only augmentation scheme; our results indicate that this form of augmentation outperforms traditional data augmentation and produces highly sample efficient estimators. We train a generic U-Net-based architecture under a curriculum consisting of two phases: initially relaxing the landmarking task by enlarging the label points to regions, then gradually eroding these label regions back to the base task. We measure the benefits of this approach on six datasets of radiographs with gold-standard expert annotations.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **the accuracy and efficiency of automatically detecting medical landmarks in hip X - ray images**. Specifically, the author focuses on how to predict clinical landmarks in hip X - ray images through automated methods to assist orthopedic surgeons in evaluating total hip arthroplasty (THA) before, during, and after surgery. These landmarks are crucial for measuring key parameters such as pelvic tilt angle and implant cup tilt angle, and these parameters directly affect the success rate of the surgery and the patient's recovery. ### Main Challenges 1. **Limitations of Manual Labeling**: Currently, orthopedic surgeons need to manually select these landmarks for evaluation, which is a time - consuming and error - prone process. 2. **Traditional Methods of Data Augmentation Are Not Applicable**: Traditional image data augmentation methods (such as rotation, flipping, scaling, etc.) will destroy the consistency and accuracy of medical images, causing the labels to become invalid. 3. **Low Sample Efficiency**: Traditional methods require a large amount of labeled data to achieve better performance, and high - quality medical image labeling is costly and difficult to obtain. ### Solutions To solve the above problems, the author proposes a new **Label Augmentation Method**. This method performs dilation and erosion operations on the labels instead of directly enhancing the images. This method can improve the sample efficiency of the model and avoid the negative impacts brought by traditional data augmentation methods. The specific steps are as follows: 1. **Label Dilation**: Expand the original label points into regions, making the task easier in the early training stage. 2. **Gradual Erosion**: As the training progresses, gradually shrink these dilated label regions and finally return to the original task. 3. **Dynamic Weight Adjustment**: To address the label imbalance problem, a dynamic weight adjustment mechanism is introduced to ensure that each label maintains a relatively consistent loss value throughout the learning process. ### Experimental Results The author conducted experiments on six different hip X - ray datasets, and the results show that: - The proposed method is significantly superior to the baseline method, and the error distance is reduced from more than 100 pixels to 1 - 4 pixels. - Traditional data augmentation methods perform poorly and are even harmful in this type of task. - On some labels, although the size of the used dataset is small, the proposed model still exceeds the best results in the existing literature. ### Conclusion This study demonstrates the effectiveness of the label augmentation method in automatically detecting landmarks in medical images, especially in hip X - ray images. Future work will further collect test datasets and explore the potential applications of this method in other medical fields.

Label Augmentation Method for Medical Landmark Detection in Hip Radiograph Images

Improving Image Classification of Knee Radiographs: An Automated Image Labeling Approach

A universal medically assisted model for anatomical landmark detection in radioactive images

Semi-supervised Anatomical Landmark Detection via Shape-regulated Self-training

Anatomic-landmark Detection Using Graphical Context Modelling.

Automated Cephalometric Landmark Detection: A Novel Software Model Compared with Manual Annotation Method

Automatic Location and Semantic Labeling of Landmarks on 3D Human Body Models

A Cascade Regression Model For Anatomical Landmark Detection

Learn Fine-Grained Adaptive Loss for Multiple Anatomical Landmark Detection in Medical Images

LandmarkNet: a 2D Digital Radiograph Landmark Estimator for Registration

Medical Data Augmentation Using Generative Adversarial Networks : X-ray Image Generation for Transfer Learning of Hip Fracture Detection

Segmentation-Guided Coordinate Regression for Robust Landmark Detection on X-Rays: Application to Automated Assessment of Lower Limb Alignment

UNSCT-HRNet: Modeling Anatomical Uncertainty for Landmark Detection in Total Hip Arthroplasty

An Automatic Cephalometric Landmark Detection Method Based on Heatmap Regression and Monte Carlo Dropout

CircleNet for Hip Landmark Detection

Multi-target landmark detection with incomplete images via reinforcement learning and shape prior embedding

CAMLD: Contrast-Agnostic Medical Landmark Detection with Consistency-Based Regularization

Robust Multi-scale Anatomical Landmark Detection in Incomplete 3D-CT Data

One-Shot Medical Landmark Localization by Edge-Guided Transform and Noisy Landmark Refinement

SIGACT News Online Algorithms Column 29

CNN-based Landmark Detection in Cardiac CTA Scans