Prior information-based high-resolution tomography image reconstruction from a single digitally reconstructed...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Computed tomography (CT) scanning is vital for clinical diagnosis and surgery.
• High radiation exposure from conventional CT poses health risks.
• Reconstructing images from limited data is a key challenge.

Purpose of the Study:

• To develop a deep learning framework for efficient 3D tomography image reconstruction from sparse views.
• To reduce radiation dose associated with CT scans.
• To enhance the quality and detail of reconstructed images.

Main Methods:

• A novel deep learning framework, TIReconNet, was developed.
• The framework integrates feature extraction, shape mapping, volume generation, and super-resolution modules.
• It combines 2D and 3D operations for efficient processing.

Main Results:

• TIReconNet successfully reconstructed high-resolution 3D tomography images.
• The method achieved a PSNR of 18.621 ± 1.228 dB and SSIM of 0.872 ± 0.041 on chest radiographs.
• The framework demonstrated efficient processing with maintained spatial information.

Conclusions:

• Deep learning can generate high-resolution 3D tomographic images from sparse 2D data.
• TIReconNet offers a promising solution for radiation reduction in medical imaging.
• The technology has potential applications in trauma surgery and orthopedic guidance.