Fast SPECT/CT planar bone imaging enabled by deep learning enhancement.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Nuclear Medicine

Background:

• Deep learning shows promise for reducing SPECT scan durations.
• Previous models used simulated data; real-world validation is limited due to scan misalignment.
• Accelerating bone scintigraphy is crucial for patient comfort and throughput.

Purpose of the Study:

• To develop a deep learning method for generating high-quality whole-body bone images from 2x and 3x accelerated scans.
• To validate the method's effectiveness on real clinical data.

Main Methods:

• A prospective study of 76 patients undergoing standard and accelerated (2x, 3x) whole-body bone scans.
• Utilized a content-attention image restoration approach based on Residual-in-Residual Dense Block (RRDB).
• Evaluated image quality using Learned Perceptual Image Patch Similarity (LPIPS), Fréchet Inception Distance (FID), and expert physician review.

Main Results:

• The method achieved state-of-the-art performance in FID and LPIPS for both 2x and 3x fast scans.
• Restored images showed significant improvements in quality, Tc-99m MDP distribution, and artifact reduction compared to fast scans.
• Nuclear physicians confirmed enhanced diagnostic confidence in the accelerated images.

Conclusions:

• The deep learning approach effectively enhances image quality for accelerated whole-body bone scans using real clinical data.
• The method demonstrates robustness to misalignment and improves diagnostic value.
• This technique holds potential for efficient, high-quality fast bone imaging in clinical settings.