Quantitative evaluation of unsupervised clustering algorithms for dynamic total-body PET image analysis
Oona Rainio1, Maria K Jaakkola1, Riku Klén1
1Turku PET Centre, University of Turku and Turku University Hospital, Turku, Finland.
Journal of Medical Engineering & Technology
|February 20, 2025
Summary
Gaussian mixture model (GMM), fuzzy c-means (FCM), and independent component analysis (ICA) with mini batch K-means show promise for dynamic total-body PET analysis, achieving high accuracy in classifying time activity curves.
Area of Science:
- Medical Imaging
- Positron Emission Tomography (PET)
- Data Analysis
Background:
- Dynamic total-body PET imaging is a recent advancement enabled by new scanner technology.
- Systematic evaluation of clustering algorithms for processing dynamic total-body PET data is limited.
Purpose of the Study:
- To compare the performance of 15 unsupervised clustering methods for processing dynamic total-body PET images.
- To identify effective algorithms for classifying time activity curves (TACs) in dynamic PET data.
Main Methods:
- Compared 15 unsupervised clustering algorithms on dynamic total-body 15O-water PET images from 30 patients.
- Utilized K-means, PCA, ICA, GMM, FCM, agglomerative clustering, and spectral clustering.
- Classified 5000 TACs per image from specific organs and tissues to evaluate accuracy.
Main Results:
- Gaussian mixture model (GMM) achieved 89% median accuracy.
- Fuzzy c-means (FCM) achieved 83% median accuracy.
- ICA combined with mini batch K-means achieved 81% median accuracy, with processing times under 0.5 seconds.
Conclusions:
- GMM, FCM, and ICA with mini batch K-means are promising for dynamic total-body PET analysis.
- These methods offer efficient and accurate classification of TACs.
- Further research into these algorithms can advance dynamic total-body PET data processing.
