|
Abstract
|
Purpose
This review aims to investigate the available evidence of PET image reconstruction using conventional and AI-based approaches.
Materials and methods
The electronic literature search was conducted in the PubMed and Scopus database for English articles published up to November 30, 2021.
Results
Positron emission tomography (PET) is a nuclear imaging modality that uses radioactive material to measure metabolic activity. An influencing and important factor for PET images is the image reconstruction algorithm. Image reconstruction approaches use the raw data to produce an accurate and meaningful activity distribution. In recent years, many efforts have been done to produce high-quality PET images by using analytical reconstruction algorithms, the combination of computed tomography (CT) or multimodal Magnetic Resonance Imaging (MRI), and AI-based architecture. In this paper, we first review the conventional and AI-based PET image reconstruction approaches. Next, some criteria for assessment of the quality of PET images are introduced. Finally, different AI-based PET image reconstruction approaches are compared.
Conclusion
If the training dataset is too small or not representative, the resulting model will be compromised. Based on the obtained results of research about PET image reconstruction, using big and more representative datasets, the AI-based approaches can go beyond conventional PET image reconstruction algorithms. But the main problem of AI-based algorithms is that clinical validation and adoption of these tools face many challenges. Also, the acquisition of a task-based dataset can be promising for improving the performance of AI-based PET image reconstruction.
|