amyloidAI

Estimation of brain amyloid accumulation using deep learning in clinical [¹¹C]PiB PET imaging.

Cite:

When using, please cite

Ladefoged, C.N., Anderberg, L., Madsen, K. et al.
Estimation of brain amyloid accumulation using deep learning in clinical [11C]PiB PET imaging.
EJNMMI Phys 10, 44 (2023).
https://doi.org/10.1186/s40658-023-00562-7

Usage:

The algorithms require the PET image to be skull stripped and resampled to MNI space in 1x1x1 mm³ resolution and 256x256x256 matrix.

The templates to do this guided by CT or MRI are supplied. If the PET image is already preprocessed, simply call:

amyloidAI -i <PET.nii.gz>

To run the preprocessing as part of the algorithm you need to supply a CT or an MRI image that is aligned with the PET image:

amyloidAI -i <PET.nii.gz> --CT <CT.nii.gz>

or

amyloidAI -i <PET.nii.gz> --MRI <MRI.nii.gz>

Algorithm output and interpretation

The algorithm outputs (and prints) a dictionary with the keys: suvr, suvr_std, diagnosis, diagnosis_std. The main output, suvr and diagnosis, is the median and mean, respectively, of the 5-fold ensemble inference. The *_std outputs are the standard deviation of the five predicted values.

SUVr

SUVr < 1.35 can be interpreted as amyloid negative and > 1.35 as amyloid positive. A high standard deviation indicates less model certainty for the diagnosis of the patient.

Diagnosis

A diagnosis value < 0.5 indicates amyloid negative and > 0.5 amyloid positive. Here too, a larger standard deviation indicates less certainty for the diagnosis.