Research map / Image generation / Image restoration
A Deep Step Pattern Representation for Multimodal Retinal Image Registration
Jimmy Addison Lee, Peng Liu, Jun Cheng, Huazhu Fu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train on mixed multimodal data instead of best single-modality model (ablation) | DeepSPa CNN · Patch classification, total (no thresholding) | accuracy (%): 63.10 -> 86.04 | Improved |
| train on mixed multimodal data instead of best single-modality model (ablation) | DeepSPa CNN · Patch classification, total (threshold 0.68) | accuracy (%): 71.13 -> 90.11 | Improved |
| add confidence thresholding at 0.68 (ablation) | DeepSPa multimodal CNN · Patch classification, total | accuracy (%): 86.04 -> 90.11 | Improved |
| add confidence thresholding at 0.68 (ablation) | FA-only CNN · Patch classification, total | accuracy (%): 59.19 -> 65.61 | Improved |
| raise classification confidence threshold (ablation) | DeepSPa CNN · Patch classification | accuracy and remaining features: accuracy increases, remaining features drop | Mixed |
| use DeepSPa learned step patterns instead of LoSPA hand-designed patterns | registration pipeline · Color fundus-FA (mild-to-moderate) | success rate (%): 93.33 -> 96.67 | Improved |
| use DeepSPa learned step patterns instead of LoSPA hand-designed patterns | registration pipeline · Color fundus-FA (severe) | success rate (%): 79.17 -> 86.5 | Improved |
| use DeepSPa learned step patterns instead of LoSPA hand-designed patterns | registration pipeline · Color fundus-OCT | success rate (%): 63.75 -> 81.25 | Improved |
| use DeepSPa instead of LoSPA | registration pipeline · Color fundus-OCT | average RMSE (pixels): 3.43 -> 2.62 | Improved |
| use affine instead of second-order polynomial transformation (ablation) | DeepSPa · Retinal registration | registration accuracy | No clear effect |
| rotate floating image from 0 to 180 degrees | DeepSPa · 20 multimodal pairs, rotation test | registration success: all pairs registered at 0 to 180 degrees | No clear effect |
| rescale floating image up to factor 2.8 | DeepSPa · 20 multimodal pairs, scale test | registration success: robust at 1.8 and below, usually fails above 1.9 | Hurt |
Closest papers by meaning
- SuperJunction: Learning-Based Junction Detection for Retinal Image Registration (AAAI 2024 · similarity 0.85)
- FOCUS: Benchmarking Retinal Model Generalization from Foundation Vision Encoders to Multimodal LLMs (NeurIPS 2026 · similarity 0.84)
- Fast Symmetric Diffeomorphic Image Registration with Convolutional Neural Networks (CVPR 2020 · similarity 0.83)
- Task Augmentation-Based Meta-Learning Segmentation Method for Retinopathy (TPAMI 2025 · similarity 0.83)
- SACB-Net: Spatial-awareness Convolutions for Medical Image Registration (CVPR 2025 · similarity 0.82)
- Deep Learning in Medical Image Registration: Magic or Mirage? (NeurIPS 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
