Simultaneous attenuation and scatter correction of PET data in the image: quantitative and clinical assessment of image-to-image deep learning models.
Authors
Affiliations (4)
Affiliations (4)
- Department of Medical Physics and Biomedical Engineering, Tehran University of Medical Sciences, Tehran, Iran; Research Center for Molecular and Cellular Imaging (RCMCI), Advanced Medical Technologies and Equipment Institute (AMTEI), Tehran University of Medical Sciences, Tehran, Iran.
- Chronic Respiratory Diseases Research Center, National Research Institute of Tuberculosis and Lung Diseases (NRITLD), Shahid Beheshti University of Medical Sciences, Tehran, Iran; PET/CT and Cyclotron Center, Masih Daneshvari Hospital, Shahid Beheshti University of Medical Sciences, Tehran, Iran.
- Division of Nuclear Medicine and Molecular Imaging, Geneva University Hospital, CH-1211 Geneva, Switzerland.
- Department of Medical Physics and Biomedical Engineering, Tehran University of Medical Sciences, Tehran, Iran; Research Center for Molecular and Cellular Imaging (RCMCI), Advanced Medical Technologies and Equipment Institute (AMTEI), Tehran University of Medical Sciences, Tehran, Iran. Electronic address: [email protected].
Abstract
Positron Emission Tomography (PET) non-invasively assesses body metabolism but requires correction for photon attenuation and scatter. This study aims to convert non-attenuation scatter corrected (NASC) brain PET/CT images to measured attenuation and scatter corrected (MASC) images using deep learning algorithms. In this study, brain PET/CT (<sup>18</sup>F-FDG) images of 125 patients diagnosed with epilepsy disorder were used. Two convolutional neural networks, UNET and CGAN, were implemented to take 2D-NASC inputs and generate MASC images. For a precise evaluation, segmentation of 83 brain regions and the calculation of 19 radiomic features for each region were performed, and paired sample t-test was conducted for each region to determine significant differences between the predicted and ground-truth images. Clinical evaluation was performed by nuclear medicine physicians. The mean Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM), and Root Mean Square Error (RMSE) were 41.55 ± 3.55 (dB), 0.997 ± 0.002, and 294.8 ± 131.0 (Bq/ml) for the CGAN model, and 40.70 ± 3.84 (dB), 0.996 ± 0.003, and 327.6 ± 163.8 (Bq/ml) for the UNET model, respectively. Among 1,577 paired values of radiomic features in different brain regions of test set, only 126 (CGAN) and 156 (UNET) values showed significant differences from the corresponding MASC values, indicating that the majority of radiomic features were well preserved while still highlighting subtle textural variations. In clinical evaluation, the UNET and CGAN models demonstrated strong performance in visual quality assessment. Direct attenuation and scatter correction using both convolutional deep networks is a promising approach for brain PET images, when CT is not available.