Related Experiment Video
Updated: Jun 10, 2025
Computed Tomography-guided Time-domain Diffuse Fluorescence Tomography in Small Animals for Localization of Cancer Biomarkers
Published on: July 17, 2012
Deep-learning-based attenuation map generation in kidney single photon emission computed tomography.
Kyounghyoun Kwon1,2, Dongkyu Oh2,3, Ji Hye Kim2
1Department of Health Science and Technology, Graduate School of Convergence Science and Technology, Seoul National University, Gwanggyo-ro 145, Yeongtong-gu, Suwon, Gyeonggi-do, 16229, Republic of Korea.
This study developed an AI algorithm to create synthetic attenuation maps for kidney SPECT imaging, eliminating the need for CT scans and reducing radiation exposure. This enables CT-free SPECT imaging for accurate GFR measurement, improving patient safety.
Area of Science:
- Nuclear Medicine
- Medical Imaging
- Artificial Intelligence
Background:
- Accurate attenuation correction (AC) is crucial for quantitative SPECT/CT imaging.
- CT scans increase radiation exposure and are essential for AC.
- Developing CT-free quantification methods is vital for nuclear medicine.
Purpose of the Study:
- To establish a CT-free quantification technology in kidney SPECT imaging.
- To generate synthetic attenuation maps (μ-maps) from SPECT data using deep learning.
- To reduce radiation exposure and eliminate the need for CT scans in SPECT imaging.
Main Methods:
- A modified 3D U-Net deep learning model was trained on 800 Tc-99m DTPA SPECT/CT scans.
- Investigated the impact of SPECT data types, normalization, loss functions, and interpolation on μ-map generation.
- Evaluated checkerboard artifacts and the influence of iodine contrast media.
Main Results:
- Optimized AI model using scattering SPECT, logarithmic maximum normalization, L1 + 3xLGDL loss, and nearest-neighbor interpolation significantly improved μ-map generation (p < 0.00001).
- Nearest-neighbor interpolation effectively eliminated checkerboard artifacts.
- AI-generated μ-maps were neutral to iodine contrast, showing negligible effects on GFR measurements.
- Potential radiation dose reduction ranged from 45.3% to 78.8%.
Conclusions:
- Successfully developed and optimized a deep learning algorithm for synthetic μ-map generation in kidney SPECT.
- Demonstrated the feasibility of transitioning from SPECT/CT to CT-free SPECT imaging for GFR measurement.
- This advancement enhances patient safety and efficiency in nuclear medicine.
