Pixel contrastive learning improves spine landmark detection on X-rays
Original title: Dense Representation Learning and Differentiable Shape Modeling for Spine Landmark Detection
A U-Net matches dense pixel embeddings to landmark prototypes, then uses learnable shape priors to resolve ambiguous vertebral corners.
Researchers reframe spinal X-ray landmark detection as pixel-level contrastive learning that matches dense embeddings with landmark prototypes. With a generic U-Net, the method produces more discriminative heatmaps than traditional heatmap regression and separates corners on different vertebrae. Learnable shape priors over the global landmark layout resolve remaining location ambiguity. The authors report competitive landmark accuracy and clinical parameter estimates on the public AASCE dataset and two private X-ray sets, and release code and weights.
- Heatmaps are combined with learnable priors over the full spinal landmark layout.
- Evaluation covers AASCE plus two private spinal X-ray datasets.
- Code and pre-trained weights are at github.com/BioMedTP/Pix2Point.
From the original article
1 min read at arxiv.orgSpine landmark detection from spinal X-ray is a key step for estimating spine deformation severity, for building biomechanical models, and for pre-operative planning. We reframe anatomical landmark detection as a pixel-level contrastive representation learning problem, matching dense pixel embeddings with landmark prototypes.
By Augustin Daridon · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.