DX2CT: Diffusion Model for 3D CT Reconstruction from Bi or Mono-planar 2D X-ray(s)

Abstract: Computational tomography (CT) provides high-resolution medical imaging, but it can expose patients to high radiation. X-ray scanners have low radiation exposure, but their resolutions are low. This paper proposes a new conditional diffusion model, DX2CT, that reconstructs three-dimensional (3D) CT volumes from bi or mono-planar X-ray image(s). Proposed DX2CT consists of two key components: 1) modulating feature maps extracted from two-dimensional (2D) X-ray(s) with 3D positions of CT volume using a new transformer and 2) effectively using the modulated 3D position-aware feature maps as conditions of DX2CT. In particular, the proposed transformer can provide conditions with rich information of a target CT slice to the conditional diffusion model, enabling high-quality CT reconstruction. Our experiments with the bi or mono-planar X-ray(s) benchmark datasets show that proposed DX2CT outperforms several state-of-the-art methods. Our codes and model will be available at: https://www.github.com/intyeger/DX2CT.

• The paper presents DX2CT, a novel conditional diffusion model that reconstructs 3D CT volumes from bi or mono-planar X-ray images using feature map modulation and position-aware conditioning.
• It employs a transformer to integrate 2D X-ray features with 3D positional information, enabling robust CT reconstruction with minimal input data.
• Experimental results show that DX2CT outperforms state-of-the-art methods by delivering superior image quality and efficiency, offering a safer alternative to traditional CT scans.

The paper "DX2CT: Diffusion Model for 3D CT Reconstruction from Bi or Mono-planar 2D X-ray(s)" (2409.08850) introduces a novel method for reconstructing three-dimensional (3D) computed tomography (CT) volumes using bi or mono-planar two-dimensional (2D) X-ray images. This significant advancement aims to bridge the gap between the high resolution yet high radiation exposure of CT scans and the lower resolution but safer X-ray imaging.

The method, DX2CT, stands out due to its innovative use of a conditional diffusion model with two critical components:

1. Feature Map Modulation: Utilizing a new transformer, the feature maps extracted from 2D X-rays are modulated with 3D positions of the CT volume. This allows for a rich transfer of information about the target CT slice.
2. Position-aware Feature Maps: These modulated 3D position-aware feature maps effectively condition the diffusion model, enabling high-quality CT reconstructions.

Experiments on benchmark datasets demonstrate that DX2CT outperforms other state-of-the-art methods, delivering superior reconstructions from fewer input images.

For context, the field of low-dose CT reconstruction has seen various approaches over the years. For example, a deep convolutional neural network (CNN) using directional wavelets was proposed for reducing noise and artifacts in low-dose CT images (Kang et al., 2016). This method efficiently processes wavelet transform coefficients to enhance image quality.

Moreover, recent advancements such as the "Improving 3D Imaging with Pre-Trained Perpendicular 2D Diffusion Models" (Lee et al., 2023) and "Solving 3D Inverse Problems using Pre-trained 2D Diffusion Models" (Chung et al., 2022), illustrate innovative techniques where pre-trained 2D models are applied to 3D reconstruction tasks, showcasing the growing effectiveness of 2D-to-3D transformations in medical imaging.

Additionally, methods like MedNeRF leverage neural radiance fields to reconstruct 3D CT projections from single X-rays, further highlighting the potential and diversity of deep learning approaches in achieving high-fidelity 3D reconstructions (Corona-Figueroa et al., 2022).

Compared to these methods, DX2CT uniquely maximizes the information from minimal X-ray inputs through advanced conditional diffusion techniques, setting a new bar for efficiency and quality in 3D medical image reconstructions. This not only significantly reduces patient exposure to harmful radiation but also provides a cost-effective and precise alternative to traditional CT scans.