📢📢📢To use DCCCSlicer, you need to download from our Release Page.
An open-source, super-simple, ultra-fast, fully-automated, fairly-accurate and PET-only solution to conduct spatial normalization and semi-quantification for almost any brain PET modalities.
Abeta, tau, FDG, DAT, MET, synaptic density... you name it!
Use DCCC to calculate Centiloid and CenTauRz.
Supplementary.video.1.-.mosaic.mp4
Shiny new features DCCCSlicer can interpret input Abeta and tau PET images now. It not only help determine if a patient may have AD, but also identifies the regions of pathological deposition (you may need to adjust the window/level to get a better illustration).
Supplementary.video.1.mp4
- 20260208: Our paper entitled Decoupling Alzheimer’s Disease Pathology in PET with Improved Clinical Relevance via Interpretable Adversarial Decomposition Learning has just been accepted by Radiology! You can try out the
ADADscore in our 3DSlicer plugin! - 20260408: Our paper is published online by Radiolody!
It’s always a good idea to manually verify that DCCC has performed spatial normalization correctly. After processing, click the Show Normalization button to inspect the normalization quality. Poorly normalized images will result in inaccurate semi-quantitative metrics - see this issue for details. This is especially important for PET scans containing substantial non-brain anatomy (e.g., neck, shoulders). If automatic spatial normalization proves suboptimal, roll back to the original image and try one of these rescue strategies:
- Enable the
Iterative Rigidoption in the plugin interface - Or enable the
Manual FOVoption and perform a manual rigid registration (field-of-view placement). DCCC will crop this image, skip rigid registration, and go directly to Affine + Elastic normalization.
You can quickly align brain images with rigid transformation to the MNI space by localizing AC and PC. This step may be required for spatial normalization with SPM/rPOP.
localizer.demo.mp4
Relative SUV (ratio) error and time consumption on the Centiloid/CenTauRz projects.
| Methods | PiB (%) | AV45 (%) | FBB (%) | FMM (%) | NAV4694 (%) | FTP (%) | Time (s) |
|---|---|---|---|---|---|---|---|
| SPM12 | 1.33±1.07 | 1.66±1.32 | 1.40±0.85 | 3.00±2.84 | 1.90±2.77 | 1.07±1.27 | 198.96±59.37 |
| SPM PET | 7.84±5.31 | 15.75±36.98 | 14.18±11.13 | 10.70±8.41 | 16.43±9.60 | 12.40±11.39 | 6.43±1.80 |
| SPM PET (Template) | 3.97±2.85 | 6.44±3.83 | 6.16±3.27 | 8.93±3.38 | 5.43±3.27 | 3.65±2.78 | 4.41±0.96 |
| ANTs PET (Template)1 | 21.27±19.06 | 15.04±7.89 | 19.05±9.68 | 21.89±7.24 | 21.59±8.21 | 6.68±4.58 | 10.07±1.72 |
| rPOP2 | 3.13±2.37 | 5.04±8.44 | 3.51±2.74 | 4.76±4.09 | 5.72±11.29 | 5.92±5.38 | 5.32±1.05 |
| SNBPI | 2.29±1.91 | 2.24±1.97 | 2.85±2.33 | 3.83±2.98 | 3.15±2.66 | 1.41±1.13 | 160.98±46.92 |
| Ours (Pytorch) | 2.35±1.66 | 2.75±1.68 | 2.75±2.32 | 3.79±3.84 | 2.77±2.16 | 1.50±1.64 | 1.22±0.64 |
| Ours (Iterative) | 2.39±1.74 | 2.77±1.73 | 2.76±2.30 | 3.81±3.84 | 2.76±2.15 | 1.50±1.72 | 1.72±1.16 |
| Ours (ONNX) | 2.42±1.76 | 2.78±1.71 | 2.72±2.29 | 3.78±3.84 | 2.78±2.17 | 1.49±1.67 | 16.60±1.41 |
The smallest error and shortest computation time are marked with bold, while the second smallest error and shortest computation time are underlined. SPM12 is employed to reproduce the original literature results using the standard pipeline and does not participate in result ranking; it provides a baseline for reproducibility. SPM PET refers to the PET-only spatial normalization algorithm provided by SPM5, utilizing the 15O-H2O template. SPM PET (Template) refers to the same algorithm, but the templates used are the average of each tracer in the Centiloid/CenTauR dataset after normalization. Check SNBPI and rPOP for their wonderful work!
1: ANTsPy exhibited suboptimal performance on Aβ PET images, though results were more acceptable on FTP scans, possibly due to lower image quality in parts of the GAINN Centiloid Project dataset. We consulted the ANTsPy developers (issue #832), but the problem remains unresolved. Readers should note that these atypical results may not reflect ANTsPy’s performance on higher-quality Aβ PET images.
2: rPOP failed completely in spatial normalization on 3 PiB images (yielding infinite or undefined SUVr values) with its fully automated pipeline. Only valid SUVr results were included in the relative error statistics reported in the table.
Please cite these metrics/algorithms if you use them in your research
Metrics:
-
Centiloid: Klunk WE, Koeppe RA, Price JC, Benzinger TL, Devous Sr. MD, Jagust WJ, et al. The Centiloid Project: Standardizing quantitative amyloid plaque estimation by PET. Alzheimer’s & Dementia. 2015;11(1):1-15.e4. -
CenTauR: Leuzy A, Raket LL, Villemagne VL, Klein G, Tonietto M, Olafson E, et al. Harmonizing tau positron emission tomography in Alzheimer’s disease: The CenTauR scale and the joint propagation model. Alzheimer’s & Dementia. 2024;20(9):5833–48. -
CenTauRz: Villemagne VL, Leuzy A, Bohorquez SS, Bullich S, Shimada H, Rowe CC, et al. CenTauR: Toward a universal scale and masks for standardizing tau imaging studies. Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring. 2023;15(3):e12454. -
Fill States: Doering E, Hoenig MC, Giehl K, et al. “Fill States”: PET-derived Markers of the Spatial Extent of Alzheimer Disease Pathology. Radiology. 2025;314(3):e241482. doi:10.1148/radiol.241482 -
Abeta load / AmyloidIQ(working on 🚧, the following publications used the same algorithm) :- Whittington A, Sharp DJ, Gunn RN. Spatiotemporal Distribution of β-Amyloid in Alzheimer Disease Is the Result of Heterogeneous Regional Carrying Capacities. Journal of Nuclear Medicine. 2018 May 1;59(5):822–7.
- Whittington A, Gunn RN. Amyloid Load: A More Sensitive Biomarker for Amyloid Imaging. Journal of Nuclear Medicine. 2019 Apr 1;60(4):536–40.
- Rizzo G, Whittington A, Hesterman J, Gunn RN. AmyloidIQ: An advanced analytical algorithm to quantify amyloid-PET [18F]NAV4694 scans. Alzheimer’s & Dementia. 2020 Dec;16(S4):e043823.
-
Abeta index(working on 🚧, the following publications used the same algorithm) :- Lilja J, Leuzy A, Chiotis K, Savitcheva I, Sörensen J, Nordberg A. Spatial normalization of 18F-flutemetamol PET images using an adaptive principal-component template. Journal of Nuclear Medicine. 2019;60(2):285–291.
- Leuzy A, Lilja J, Buckley CJ, Ossenkoppele R, Palmqvist S, Battle M, et al. Derivation and utility of an Aβ-PET pathology accumulation index to estimate Aβ load. Neurology. 2020;95(21):e2834–e2844.
-
ADAD: not published yet
Spatial normalization algorithms:
DCCC SPM12 style: not published yetFast and Accurate(the following publications used the same algorithm):- Kang SK, Kim D, Shin SA, Kim YK, Choi H, Lee JS. Fast and Accurate Amyloid Brain PET Quantification Without MRI Using Deep Neural Networks. Journal of Nuclear Medicine. 2023 Apr 1;64(4):659–66.
- Kim D, Kang SK, Shin SA, Choi H, Lee JS. Improving 18F-FDG PET Quantification Through a Spatial Normalization Method. Journal of Nuclear Medicine. 2024 Aug 29; Available from: https://jnm.snmjournals.org/content/early/2024/08/29/jnumed.123.267360
- Kang SK, Kim D, Shin SA, Kim YK, Choi H, Lee JS. Evaluation of BTXBrain-Tau, an AI-powered automated quantification software for Flortaucipir PET images. Alzheimer’s & Dementia. 2023;19(S16):e074520.
- Yoo HB, Kang SK, Shin SA, Kim D, Choi H, Kim YK, et al. Artificial Intelligence–Powered Quantification of Flortaucipir PET for Detecting Tau Pathology. Journal of Nuclear Medicine. 2025 Sep 11; Available from: https://jnm.snmjournals.org/content/early/2025/09/11/jnumed.125.269636
Please note that Lee’s algorithm is not publicly available. The version we reproduced based on the DCCC framework may differ in network architecture and implementation details, and it has not yet undergone extensive validation. While Lee et al. achieved cross-modality applicability through transfer learning, our DCCC implementation was directly trained on more than 5,000 multimodal PET images to provide inherent multimodal support.
For users who prefer running the core calculator from the command line (including batch processing and advanced options), please refer to the standalone CLI documentation. Release assets now publish the headless package separately as DCCCcore-<version>-<platform>.zip, alongside the full DCCCSlicer extension package.
- Add support for skipping spatial normalization to directly calculate Centiloid/CenTauR.
- Support other brain PET semi-quantitative metrics, such as Z-scores and basal ganglia asymmetry index.
- Add support for “Fill States”: PET-derived Markers of the Spatial Extent of Alzheimer Disease Pathology
- Add support for other spatial normalization algorithms.
- Added support for Fast and Accurate Amyloid Brain PET Quantification Without MRI Using Deep Neural Networks
- Improve the UI.
Check our reproduction reports on fill states, Abeta load and Abeta index!
We sincerely thank the passionate and outstanding users and contributors of DCCC. Many of our contributors come from the medical community and may not be accustomed to using GitHub, so we would like to acknowledge their contributions here. Your valuable feedback has been the greatest driving force behind the continuous improvement of the project.
| Yulai Li, MD, PhD @ Xiangya Hospital, Changsha, CHINA For providing the SV2A dataset |
Chenpeng Zhang, MD, PhD @ Renji Hospital, Shanghai, CHINA For reporting rigid registration failures |
Daoyan Hu, PhD @ The Second Affiliated Hospital Zhejiang University School of Medicine, Hangzhou, CHINA For providing testing feedback with SPM |
|---|---|---|
 |
||
| Fanglan Li, MD, PhD @ West China Hospital, Chengdu, CHINA For reporting rigid registration failures |
And more... | |
 |
Feel free to get in touch if you'd like to contribute, report a bug, share a poorly spatially normalized PET image, or suggest support for a new brain PET metric - we'll be sure to credit you! |
If you find this repo helpful for your work, please cite
@article{doi:10.1148/radiol.252321,
author = {Tang, Cheng and Sun, Xun and Tang, Anqi and Ruan, Weiwei and Liu, Fang and Fang, Hanyi and Gai, Yongkang and Liang, Zhihou and Su, Ying and Wang, Xinggang and Lan, Xiaoli},
title = {Decoupling Alzheimer Disease Pathologic Abnormalities at PET with Improved Clinical Relevance by Interpretable Adversarial Decomposition Learning},
journal = {Radiology},
volume = {319},
number = {1},
pages = {e252321},
year = {2026},
doi = {10.1148/radiol.252321},
note ={PMID: 41944723},
URL = {https://doi.org/10.1148/radiol.252321},
eprint = { https://doi.org/10.1148/radiol.252321},
abstract = { Background Template-based PET metrics quantify Alzheimer disease (AD) amyloid-β (Aβ) and tau burden but compress whole-brain data into a single scalar, overlooking disease heterogeneity and sometimes causing imaging-clinical discordance. Artificial intelligence (AI) approaches capture richer patterns but often lack biologic interpretability. Purpose To develop and validate an interpretable deep-learning framework that separates AD-specific abnormalities from physiologic uptake using pathophysiologic constraints, generating a clinically meaningful AI biomarker. Materials and Methods In this retrospective study, Aβ and tau PET scans from the Alzheimer’s Disease Neuroimaging Initiative, Australian Imaging Biomarkers and Lifestyle study, Global Alzheimer’s Association Interactive Network, and the authors’ center were analyzed. An adversarial decomposition learning (ADL) network generated voxel-level pathologic maps and an AD adversarial decomposition (ADAD) score. Discriminatory performance for clinical AD versus cognitively normal individuals was evaluated using the area under the curve (AUC). Clinical relevance was assessed with cognitive, hippocampal volume, cerebrospinal fluid (CSF), and neuropathologic measures using longitudinal mixed-effects models and Spearman correlations. Results The study included 7457 Aβ PET scans from 3595 patients (median age, 71.4 years; IQR, 65.7–77.0 years; 1637 female patients) and 1894 tau PET scans from 1127 patients (median age, 72.0 years; IQR, 66.9–78.5 years; 545 female patients). External testing AUCs were 0.94 (95\% CI: 0.89, 0.98) for Aβ and 0.98 (95\% CI: 0.95, 1.00) for tau. ADL generated interpretable pathologic attribution maps that correlated with expert rankings (Aβ and tau, Spearman ρ = 0.79 and 0.63, respectively). Although Centiloid and CenTauRz showed numerically higher correlations with postmortem neuropathologic structure and stronger associations with CSF biomarkers, the ADAD score demonstrated independent baseline and longitudinal associations with cognitive outcomes and hippocampal atrophy after adjustment. Conclusion Pathophysiologic-constrained ADL provided interpretable, personalized pathologic maps and an AI-derived ADAD score that more closely linked PET pathologic abnormalities with multimodal clinical measures. © RSNA, 2026 Supplemental material is available for this article. }
}This project is open-sourced under a CC-BY-NC 4.0 license and therefore not allowed for commercial use. This project is for research only and is prohibited in clinical practice.
IMPORTANT LICENSE UPDATE: Since DCCCSlicer V2.0, we no longer allow results generated using DCCCSlicer to be published in closed-access journals. If you use our software in your research, please consider publishing your results in an open-access journal or making them publicly available from the journal press, unless you obtain our commercial use license in advance.
