BX2S-Net

BX2S-Net reconstructs three-dimensional spinal structures from bi-planar X-ray images using a convolutional neural network with dimensionally consistent encoder–decoder architecture.

Key Features:

• Dimensionally Consistent Encoder–Decoder Architecture: Implements a convolutional neural network architecture designed to reduce semantic gaps between two-dimensional and three-dimensional feature representations.
• Dimensionality Enhancement Method: Enhances feature map dimensional consistency to enable effective fusion of multi-view information from bi-planar X-ray images.
• Full-Scale Feature Attention Guidance Module: Uses a Full-Scale Feature Attention Guidance (FFAG) module to aggregate and guide image features during progressive decoding.
• Feature-Guided Progressive Decoding: Reconstructs spinal structures through a progressive decoder guided by hierarchical feature representations.
• Training Optimization Strategy: Improves reconstruction performance using class augmentation and a spatially weighted cross-entropy loss function.

Scientific Applications:

• Three-Dimensional Spine Reconstruction: Generates 3D representations of spinal anatomy from bi-planar X-ray imaging data.
• Spinal Structure Analysis: Supports quantitative analysis of vertebral geometry and spatial spinal configuration under weight-bearing conditions.
• Clinical Imaging Research: Facilitates computational analysis of spinal disorders using reconstructed three-dimensional anatomical models.

Methodology:

BX2S-Net processes bi-planar X-ray images with a convolutional neural network using a dimensionally consistent encoder–decoder architecture, dimensionality enhancement for multi-view feature fusion, and a progressive decoding framework guided by a Full-Scale Feature Attention Guidance module with spatially weighted cross-entropy loss.