Abstract
31
Objectives: PET/CT scans typically involve complex, configurable workflows and scan protocols to image different tracers in patients with different conditions. Artificial Intelligence (AI) has the potential to introduce novel approaches to improve the workflow by implementing automatic image quality checks and derivation of advanced reconstruction parameters that can be used during image formation process [1-2]. The objective of this study is to demonstrate proof of concept of this approach by using an AI system to automatically determine advanced scatter correction parameters to optimize image quality for clinical reading.
Methods: Anonymized PET images of a total of 83 patients from multiple institutions were used. The datasets included both 18F-FDG and 68Ga-PSMA scans. The image quality of high contrast PSMA studies is sometimes impacted by the choice of the scatter correction
Methods: This is often difficult to determine prospectively and requires manual interaction. Thus, in order to assist in choosing the suitable scatter correction method and simplify the workflow, a total of 249 findings have been identified to train a deep Convolution Neural Network (CNN) model to determine the best suited scatter correction parameters for a given case. Subsequently, the trained model was used to perform an image quality check step and, if image quality could be improved, additional image was reconstructed using optimized parameters suggested by the CNN model. Results: The available data was split in a training data set and testing data set (80%/20%, respectively). The proposed deep learning based AI system reached an accuracy of 96.6% in recommending scatter correction parameters that resulting in best image quality for a given case. Figure 1 illustrates a traditional workflow that includes visual inspection and optional manual interaction to select advanced reconstruction parameters. The CNN model automatically determines advanced reconstruction parameters potentially eliminating manual workflow steps while optimizing the image quality. Conclusion: The AI system has the potential for improving the efficiency of a traditional workflow by automating steps that often require user interaction. The design of the AI model can be expanded to other reconstruction parameters as well, further improving workflow and image quality. Research Support: None [1] Litjens G, Kooi T, Bejnordi BE, Setio AAA, Ciompi F, Ghafoorian M, van der Laak JA, van Ginneken B. and Sánchez CI, 2017. A survey on deep learning in medical image analysis. arXiv preprint arXiv:1702.05747. [2] Guehring J, Weale P and Zuehlsdorff S, Siemens Medical Solutions USA Inc (2013). System for dynamically improving medical image acquisition quality. U.S. 8,520,920 B2.