Abstract:Accurately detecting spine vertebrae plays a crucial role in successful orthopedic surgery. However, identifying and classifying lumbar vertebrae from arbitrary spine X-ray images remains challenging due to their similar appearance and varying sizes among individuals. In this paper, we propose a novel approach to enhance vertebrae detection accuracy by leveraging both global and local spatial relationships between neighboring vertebrae. Our method incorporates a two-stage detector architecture that captures global contextual information using an intermediate heatmap from the first stage. Additionally, we introduce a detection head in the second stage to capture local spatial information, enabling each vertebra to learn neighboring spatial details, visibility, and relative offset. During inference, we employ a fusion strategy that combines spatial offsets of neighboring vertebrae and heatmap from a conventional detection head. This enables the model to better understand relationships and dependencies between neighboring vertebrae. Furthermore, we introduce a new representation of object centers that emphasizes critical regions and strengthens the spatial priors of human spine vertebrae, resulting in an improved detection accuracy. We evaluate our method using two lumbar spine image datasets and achieve promising detection performance. Compared to the baseline, our algorithm achieves a significant improvement of 13.6% AP in the CM dataset and surpasses 6.5% and 4.8% AP in the anterior and lateral views of the BUU dataset, respectively.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the accuracy of automatic detection of lumbar vertebrae in X - ray images. Specifically, the paper points out that due to factors such as changes in the field of view in spinal X - ray images, similar visual features, and excessive interfering information, automatic identification and classification of lumbar vertebrae remain challenging. These problems lead to the subjectivity and inconsistency of traditional methods in the diagnosis and treatment of spinal diseases. Therefore, the researchers proposed a new method to improve the accuracy of lumbar vertebrae detection by using the global and local spatial relationships between adjacent vertebrae. ### Main contributions of the paper: 1. **Proposed a new two - stage method**: This method encodes spinal information by generating an intermediate heat map in the first stage and inputs it together with the original image into the detection sub - network in the second stage, thereby capturing global spatial prior information. 2. **Introduced a new detection head**: This detection head focuses on capturing local spatial information, especially predicting the spatial relationship and dependence between each vertebra and its adjacent vertebrae. 3. **Improved the center point representation method**: Based on the standard Gaussian function, a new center point representation method was constructed to represent the center position of spinal vertebrae more accurately, thereby improving the accuracy and reliability of detection. ### Method overview: - **Global spatial relationship**: Through the two - stage method, the intermediate heat map generated in the first stage encodes the global spatial information of vertebrae in the entire image. In the second stage, this heat map is fused with the original image to generate a new feature map for final vertebrae detection. - **Local spatial relationship**: In addition to the global spatial relationship, a new detection head was also introduced to predict the information of each vertebra and its adjacent vertebrae, including visibility and relative offset. This information is used to optimize the central heat map of each vertebra in the inference stage, thereby improving the detection accuracy. ### Experimental results: The paper was evaluated on two lumbar vertebrae image datasets, and the results showed that this method achieved significant performance improvements compared to baseline models (such as CenterNet, YOLOv5, and Faster R - CNN). Specifically, on the CM dataset, the average performance improvement was 13.6% AP; on the front view and side view of the BUU dataset, the improvements were 6.5% and 4.8% AP respectively. Through these innovations, this paper aims to provide a more accurate and reliable method for lumbar vertebrae detection to assist doctors in the diagnosis and treatment of spinal diseases.

Exploring Neighbor Spatial Relationships for Enhanced Lumbar Vertebrae Detection in X-ray Images

Part Affinity Fields and CoordConv for Detecting Landmarks of Lumbar Vertebrae and Sacrum in X-ray Images

Anatomical Prior-Based Vertebral Landmark Detection for Spinal Disorder Diagnosis

The Lumbar Region Localization Using Bone Anatomy Feature Graphs.

Automated vertebral landmarks and spinal curvature estimation using non-directional part affinity fields

The Improved Faster-Rcnn for Spinal Fracture Lesions Detection.

Integrating biplane information and context for spine landmark detection

Sequence based local–global information fusion framework for vertebrae detection under pathological and FOV variation challenges

VertXNet: Automatic Segmentation and Identification of Lumbar and Cervical Vertebrae from Spinal X-ray Images

Lumbar spine localisation method based on feature fusion

Automatic Lumbar Spine Measurement in CT Images

SpineCLUE: Automatic Vertebrae Identification Using Contrastive Learning and Uncertainty Estimation

Automatic Lumbar MRI Detection and Identification Based on Deep Learning.

LSRC: A Long-Short Range Context-Fusing Framework for Automatic 3D Vertebra Localization.

X23D—Intraoperative 3D Lumbar Spine Shape Reconstruction Based on Sparse Multi-View X-ray Data

Automatic Vertebra Labeling in Large-Scale 3D CT using Deep Image-to-Image Network with Message Passing and Sparsity Regularization

Accurate Scoliosis Vertebral Landmark Localization on X-ray Images Via Shape-constrained Multi-stage Cascaded CNNs.

VDVM: an Automatic Vertebrae Detection and Vertebral Segment Matching Framework for C-arm X-ray Image Identification.

Automatic detection and segmentation of lumbar vertebra from X-ray images for compression fracture evaluation

A Dual Coordinate System Vertebra Landmark Detection Network with Sparse-to-Dense Vertebral Line Interpolation

LumVertCancNet: A novel 3D lumbar vertebral body cancellous bone location and segmentation method based on hybrid Swin-transformer