Quantifying Amyloid Burden in Alzheimer's: A Reproducible Centiloid Pipeline for Analysis in Resource-Constrained Settings

Background:

Standardized amyloid PET quantification using the Centiloid scale is essential for Alzheimer's disease research and therapeutic trials. Implementations require expensive MATLAB licenses and highperformance local hardware, creating significant barriers to accessibility and multi-center harmonization.

Method:

To address this, we developed a combined, cloud-based pipeline integrating SPM12 preprocessing (via free MATLAB Runtime) with FSL6 quantification, containerized within the Neurodesk platform. This work was performed as part of the 2025 CONNExIN (Comprehensive Neuroimaging Analysis Experience In
resource-constrained Settings) Program. Two CONNExIN trainee teams independently implemented complementary approaches: Team O developed the FSL quantification pipeline, while Team A implemented SPM preprocessing with FSL quantification and replicated the FSL pipeline. Using the
GAAIN reference dataset (25 AD: AD01-AD25, 25 YC: YC101-YC125), we analyzed 50-70 minute 11C-PiB frames. SPM preprocessing included motion correction, PET-MRI registration, and spatial normalization. FSL quantification used binary masks with automated intensity normalization (threshold: 1.0-4.0 SUV). SUVR was calculated as cortical mean / reference mean. Subjects with data quality issues (n=4 AD) were excluded. Validation included correlation with GAAIN SUVRs and an inter-pipeline reproducibility assessment between independently implemented SPM and FSL quantification arms. Protocol is available on https://dx.doi.org/10.17504/protocols.io.8epv55jydv1b/v1

Result:

Automated quality control identified and corrected intensity scaling issues in 32% (8/25) of AD participants, with 4 subjects excluded from final analysis due to persistent data quality issues (Figure 1). In the validated dataset, our pipeline showed strong agreement with GAAIN reference values (FSL:
r=0.90; SPM: r=0.97; Figure 2), yielding SUVRs (AD cohort: 1.96±0.49, control cohort: 0.98±0.16, n=25; Figure 3). Team A's SPM implementation demonstrated excellent accuracy against GAAIN references (r=0.97), while reproducibility between independently implemented FSL pipelines was high (r=0.92, relative difference=+12.2%), with agreement against GAAIN showing minimal bias (Table 1). The cloud implementation enabled processing without local hardware or MATLAB license costs.

Conclusion:

We present a validated, cloud-based Centiloid pipeline that eliminates cost and hardware barriers. Developed under the CONNExIN program, this tool makes standardized amyloid PET quantification - a key molecular biomarker - accessible for resource-constrained settings, facilitating multi-center research and trial work in Alzheimer's disease.

 

References:

1. Klunk WE, Koeppe RA, Price JC, et al. The Centiloid Project: standardizing quantitative amyloid plaque estimation by PET. Alzheimer's & Dementia. 2015;11(1):1-15.e1-4.
2. GAIN (Global Alzheimer's Association Interactive Network). Centiloid Project Data. https://gaain.org/centiloid-project
3. Leal JP, et al. Amyloid and tau PET imaging. Bio-protocol. 2019;9(13):e3290.
4. Heeman F, et al. Centiloid scaling for quantification of brain amyloid with [¹⁸F]flutemetamol using multiple processing methods. EJNMMI Res. 2018;8:107.
5. Navitsky M, et al. Reproducibility of Centiloid Values in Real-World Amyloid PET Data. Alzheimers Dement. 2021.
6. Lopes Alves I, et al. Stress testing the Centiloid: Precision and variability of PET quantification of amyloid pathology. Alzheimers Dement. 2024.
7. Huang SY, et al. The Development of an Automatic Amyloid PET Quantification (AmPQ) Software with MR-based and MR-free Spatial Normalization. J Med Biol Eng. 2025;45:492-
501.
8. CONNExIN. Comprehensive Neuroimaging Analysis Experience in Resource-Constrained Settings. Alzheimers Dement. 2025;21(Suppl 8):e109800.
9. Neurodesk. Neurodesk: A flexible and scalable data analysis environment for reproducible neuroimaging. https://neurodesk.org