Direct attenuation correction of brain PET images using only emission data via a deep convolutional encoder-decoder...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Nuclear Medicine

Background:

• Attenuation correction (AC) is crucial for accurate PET image quantification.
• Traditional AC methods often rely on separate transmission scans or anatomical information, which can be challenging in certain settings like PET/MRI.

Purpose of the Study:

• To develop and evaluate a deep learning-based method for direct, emission-based attenuation correction of PET images.
• To assess the performance of the proposed method using quantitative metrics and radiomic feature analysis.

Main Methods:

• A convolutional encoder-decoder network was designed to generate attenuation-corrected (AC) PET images directly from non-attenuation-corrected (NAC) images.
• The network was trained and validated on brain PET data from 129 patients.
• Image quality was assessed using Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Metric (SSIM).
• Radiomic features from 83 brain regions were analyzed to evaluate quantification accuracy and reliability.

Main Results:

• The deep learning model achieved high image quality metrics (PSNR: 39.2 ± 3.65, SSIM: 0.989 ± 0.006) on the external validation set.
• Mean relative error (RE) for SUVmean across all regions was minimal (-0.10 ± 2.14%), with only 3 regions showing significant differences.
• SUVmax analysis showed a mean RE of -3.87 ± 2.84%, with 17 regions exhibiting significant differences.

Conclusions:

• Direct emission-based attenuation correction of PET images using deep convolutional encoder-decoder networks is a viable and promising technique.
• The method demonstrates robustness, particularly for SUVmean quantification, and has potential applications in PET/MRI and dedicated brain PET scanners.
• This deep learning approach offers a novel way to perform AC without anatomical information, enhancing PET imaging capabilities.