Generative adversarial network-based attenuation correction for ^99mTc-TRODAT-1 brain SPECT.

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Artificial Intelligence in Healthcare

Background:

• Attenuation correction (AC) is crucial for accurate quantification in dopamine transporter (DAT) SPECT imaging.
• Traditional Chang's method assumes uniform attenuation, limiting its accuracy.
• Evaluating deep learning (DL)-based AC methods against Chang's method on clinical 99mTc-TRODAT-1 brain SPECT data is essential.

Purpose of the Study:

• To compare the performance of various deep learning attenuation correction (DL-AC) approaches with Chang's method for 99mTc-TRODAT-1 brain SPECT.
• To assess the effectiveness of scanner-specific, cross-scanner, and ensemble DL-AC training strategies.
• To determine the optimal DL-AC method for improving quantification accuracy in DAT SPECT.

Main Methods:

• Retrospective analysis of 260 patient 99mTc-TRODAT-1 SPECT/CT scans from two different scanners.
• Implementation of direct (DL-AC) and indirect (DL-ACμ) deep learning attenuation correction using 3D cGAN.
• Comparison of scanner-specific, cross-scanner, and ensemble DL-AC training methods against Chang's method.

Main Results:

• All DL-AC methods outperformed Chang's method in accuracy, with DL-ACμ showing superior results.
• Scanner-specific training yielded better performance than cross-scanner or ensemble training.
• DL-ACμ demonstrated normalized mean square error (NMSE) values as low as 0.0055 ± 0.0034, significantly lower than Chang's method (0.0406 ± 0.0445).

Conclusions:

• Deep learning-based attenuation correction methods are feasible and robust for 99mTc-TRODAT-1 brain SPECT.
• DL-ACμ, particularly with scanner-specific training, offers superior accuracy compared to traditional methods.
• These findings support the adoption of DL-AC for enhanced quantitative accuracy in DAT SPECT imaging.