Deep learning-based prediction of later 13N-ammonia myocardial PET image frames from initial frames
Bibliographic record
Abstract
Dynamic Myocardial Positron Emission Tomography (PET) evaluates myocardial uptake. However, extended acquisition time during the dynamic PET can be a drawback, causing patient discomfort and potential motion artifacts. To address this, we employed deep learning (DL) techniques to predict later time frames using their initial ones. We used the dataset of 350 patients who underwent 13 N-ammonia dynamic myocardial PET scans to train three DL models (U-Net, U-Net with self-attention layers, and Attention ResNet). All networks underwent three stages of training to predict the last 10, 14, and 17 late frames, respectively, using the initial 11, 7, and 4 of initial time frames. This study evaluates the performance of those models in predicting later frames of PET images. The accuracy of the predicted time frames was assessed using quantitative metrics, including peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), multi-scale SSIM (MS-SSIM), multi-scale gradient magnitude similarity deviation (MS-GMSD), The Haar perceptual similarity index (HaarPSI) and mean squared error (MSE). Results indicate that the attention-based model consistently outperforms U-Net, especially when the number of initial frames is reduced. Attention ResNet excels, achieving higher PSNR=43.1 ± 0.7, SSIM=0.98 ± 0.003, MS-SSIM=0.96 ± 0.004, and lower MSE=0.24 ± 9e-5 and MS-GMSD=0.035 ± 0.004. The importance of an adequate number of initial frames for accurate predictions is highlighted. Evaluation parameter curves further illustrate the models’ performance, showcasing the robustness of the Attention ResNet model, making it a strong candidate for time frame prediction in PET imaging. In conclusion, the results demonstrate the potential of DL models for predicting late frames of dynamic PET.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".