Image-Condition-Controlled Diffusion Network with Large-Scale CT Pretraining for CT-Free Attenuation Correction in Dynamic Cardiac PET
Organizations: Department of Radiology, Northwestern University, Chicago, IL 60611 USA · Department of Electrical and Computer Engineering, Northwestern University, Evanston, IL 60208 USA · Division of Cardiology, Department of Medicine, Northwestern University, Chicago, IL 60611 USA
Abstract
Positron emission tomography (PET) provides quantitative information on myocardial perfusion. In dynamic PET, tracer kinetics are tracked through a series of reconstructed time frames. Reliable attenuation correction is essential for accurate measurements. Clinical attenuation maps (-maps) are typically derived from computed tomography (CT), which adds radiation exposure and may introduce PET-CT misregistration due to respiratory and cardiac motion. Moreover, supervised CT-free methods rely on paired non-attenuation-corrected (NAC) PET and CT data, which are scarce, whereas large-scale unpaired CT data are more readily available. To address these challenges, we propose Large-scale Unpaired CT-Informed Diffusion (LUCID), a framework that transfers anatomical priors learned from 10,000 chest CT volumes in the public CT-RATE dataset to -map synthesis for dynamic cardiac Rb PET. First, a convolutional neural network (CNN) predicts an initial -map from each NAC PET frame. Second, the CT-pretrained diffusion backbone refines a low-noise perturbation of the CNN prediction through a single denoising step, conditioned on the corresponding NAC PET frame via ControlNet. Finally, frame-wise predictions are averaged voxel-wise to obtain the final -map. We further introduce the voxel-wise temporal standard deviation (TSD) to quantify inter-frame variability without requiring a reference -map. We evaluated LUCID on an in-house dataset of 102 patients and found that it achieved the best quantitative accuracy while maintaining competitive inter-frame consistency. Ablation studies showed that refinement with a diffusion backbone pretrained on in-house CT data from the training patients was less accurate than the CNN prediction alone, highlighting the importance of large-scale CT pretraining.
Figures & tables
| Method | Rest | Stress | ||||||
|---|---|---|---|---|---|---|---|---|
| MAE | PSNR | SSIM | TSD | MAE | PSNR | SSIM | TSD | |
| 3D UNet [ 33 ] | 1.52 0.85 ∗ | 27.58 2.89 ∗ | 0.938 0.039 ∗ | 0.500 0.280 † | 1.63 0.88 ∗ | 26.85 2.74 ∗ | 0.932 0.041 ∗ | 0.559 0.298 † |
| Attn Pix2Pix [ 22 ] | 1.49 0.70 ∗ | 27.71 2.48 ∗ | 0.935 0.037 ∗ | 0.617 0.302 ∗ | 1.62 0.73 ∗ | 26.94 2.36 ∗ | 0.929 0.039 ∗ | 0.691 0.323 ∗ |
| BBDM [ 38 ] | 2.25 1.25 ∗ | 25.50 3.70 ∗ | 0.915 0.042 ∗ | 0.682 0.480 | 2.42 1.24 ∗ | 24.70 3.27 ∗ | 0.907 0.043 ∗ | 0.800 0.589 |
| DDPM [ 31 ] | 2.34 0.91 ∗ | 26.36 2.71 ∗ | 0.886 0.045 ∗ | 0.490 0.257 | 2.44 0.93 ∗ | 25.79 2.46 ∗ | 0.883 0.046 ∗ | 0.574 0.265 |
| Ours | 1.37 0.77 | 28.44 2.87 | 0.944 0.036 | 0.515 0.278 | 1.49 0.78 | 27.63 2.65 | 0.939 0.037 | 0.578 0.283 |
| Frame usage | Method | Rest | Stress | ||||||
|---|---|---|---|---|---|---|---|---|---|
| MAE | PSNR | SSIM | TSD | MAE | PSNR | SSIM | TSD | ||
| Single frame | w/o Refinement | 1.49 0.82 ∗ | 27.41 2.75 ∗ | 0.935 0.041 ∗ | – | 1.61 0.78 ∗ | 26.60 2.53 ∗ | 0.930 0.039 ∗ | – |
| w/o Prior | 2.19 1.09 ∗ | 25.33 2.56 ∗ | 0.907 0.050 ∗ | – | 2.29 1.13 ∗ | 24.89 2.43 ∗ | 0.903 0.049 ∗ | – | |
| Ours | 1.46 0.81 ∗ | 27.64 2.82 ∗ | 0.938 0.040 ∗ | – | 1.58 0.77 ∗ | 26.82 2.58 ∗ | 0.933 0.038 ∗ | – | |
| Input averaging | w/o Refinement | 1.38 0.82 | 28.13 2.94 ∗ | 0.942 0.038 ∗ | – | 1.51 0.84 | 27.29 2.80 ∗ | 0.936 0.039 ∗ | – |
| w/o Prior | 2.04 0.98 ∗ | 25.92 2.51 ∗ | 0.914 0.044 ∗ | – | 2.17 1.02 ∗ | 25.36 2.37 ∗ | 0.908 0.046 ∗ | – | |
| Method | Pretraining | Rest | Stress | ||||||
|---|---|---|---|---|---|---|---|---|---|
| MAE | PSNR | SSIM | TSD | MAE | PSNR | SSIM | TSD | ||
| CNN prior | – | 1.39 0.77 | 28.30 2.81 | 0.942 0.037 | 0.558 0.290 | 1.50 0.78 | 27.52 2.62 | 0.937 0.038 | 0.621 0.294 |
| Pure-noise | In-house | 2.35 0.87 ∗ | 26.13 2.29 | 0.889 0.042 ∗ | 0.530 0.264 | 2.46 0.92 ∗ | 25.60 2.20 | 0.885 0.044 ∗ | 0.613 0.320 |
| CT-RATE | 2.10 1.04 | 25.98 2.42 | 0.911 0.047 | 0.585 0.399 | 2.25 1.12 | 25.42 2.41 | 0.906 0.048 | 0.659 0.464 | |
| Ours | In-house | 1.46 0.76 ∗ | 28.38 2.85 ∗ | 0.941 0.037 ∗ | 0.530 0.281 ∗ | 1.58 0.76 ∗ | 27.55 2.62 ∗ | 0.936 0.037 ∗ | 0.592 0.285 ∗ |
| CT-RATE | 1.37 0.77 | 28.44 2.87 | 0.944 0.036 | 0.515 0.278 | 1.49 0.78 | 27.63 2.65 | 0.939 0.037 | 0.578 0.283 | |
PET/CT Imaging
Positron emission tomography/computed tomography (PET/CT) imaging combines metabolic and anatomical information for diagnosis, treatment planning, and disease monitoring.
See all 54 papers