Yan-Ran Joyce Wang1, Lucia Baratto1, K Elizabeth Hawk1, Ashok J Theruvath1, Allison Pribnow2, Avnesh S Thakor1, Sergios Gatidis3, Rong Lu4, Santosh E Gummidipundi4, Jordi Garcia-Diaz1, Daniel Rubin5,6, Heike E Daldrup-Link7,8. 1. Department of Radiology, Molecular Imaging Program at Stanford, Stanford University, 725 Welch Road, CA, 94304, Stanford, USA. 2. Department of Pediatrics, Pediatric Oncology, Lucile Packard Children's Hospital, Stanford University, Stanford, CA, 94304, USA. 3. Department of Diagnostic and Interventional Radiology, University Hospital Tuebingen, Tuebingen, Germany. 4. Quantitative Sciences Unit, School of Medicine, Stanford University, Stanford, CA, 94304, USA. 5. Department of Radiology, Molecular Imaging Program at Stanford, Stanford University, 725 Welch Road, CA, 94304, Stanford, USA. drubin@stanford.edu. 6. Department of Pediatrics, Pediatric Oncology, Lucile Packard Children's Hospital, Stanford University, Stanford, CA, 94304, USA. drubin@stanford.edu. 7. Department of Radiology, Molecular Imaging Program at Stanford, Stanford University, 725 Welch Road, CA, 94304, Stanford, USA. heiked@stanford.edu. 8. Department of Pediatrics, Pediatric Oncology, Lucile Packard Children's Hospital, Stanford University, Stanford, CA, 94304, USA. heiked@stanford.edu.
Abstract
PURPOSE: To generate diagnostic 18F-FDG PET images of pediatric cancer patients from ultra-low-dose 18F-FDG PET input images, using a novel artificial intelligence (AI) algorithm. METHODS: We used whole-body 18F-FDG-PET/MRI scans of 33 children and young adults with lymphoma (3-30 years) to develop a convolutional neural network (CNN), which combines inputs from simulated 6.25% ultra-low-dose 18F-FDG PET scans and simultaneously acquired MRI scans to produce a standard-dose 18F-FDG PET scan. The image quality of ultra-low-dose PET scans, AI-augmented PET scans, and clinical standard PET scans was evaluated by traditional metrics in computer vision and by expert radiologists and nuclear medicine physicians, using Wilcoxon signed-rank tests and weighted kappa statistics. RESULTS: The peak signal-to-noise ratio and structural similarity index were significantly higher, and the normalized root-mean-square error was significantly lower on the AI-reconstructed PET images compared to simulated 6.25% dose images (p < 0.001). Compared to the ground-truth standard-dose PET, SUVmax values of tumors and reference tissues were significantly higher on the simulated 6.25% ultra-low-dose PET scans as a result of image noise. After the CNN augmentation, the SUVmax values were recovered to values similar to the standard-dose PET. Quantitative measures of the readers' diagnostic confidence demonstrated significantly higher agreement between standard clinical scans and AI-reconstructed PET scans (kappa = 0.942) than 6.25% dose scans (kappa = 0.650). CONCLUSIONS: Our CNN model could generate simulated clinical standard 18F-FDG PET images from ultra-low-dose inputs, while maintaining clinically relevant information in terms of diagnostic accuracy and quantitative SUV measurements.
PURPOSE: To generate diagnostic 18F-FDG PET images of pediatric cancerpatients from ultra-low-dose 18F-FDG PET input images, using a novel artificial intelligence (AI) algorithm. METHODS: We used whole-body 18F-FDG-PET/MRI scans of 33 children and young adults with lymphoma (3-30 years) to develop a convolutional neural network (CNN), which combines inputs from simulated 6.25% ultra-low-dose 18F-FDG PET scans and simultaneously acquired MRI scans to produce a standard-dose 18F-FDG PET scan. The image quality of ultra-low-dose PET scans, AI-augmented PET scans, and clinical standard PET scans was evaluated by traditional metrics in computer vision and by expert radiologists and nuclear medicine physicians, using Wilcoxon signed-rank tests and weighted kappa statistics. RESULTS: The peak signal-to-noise ratio and structural similarity index were significantly higher, and the normalized root-mean-square error was significantly lower on the AI-reconstructed PET images compared to simulated 6.25% dose images (p < 0.001). Compared to the ground-truth standard-dose PET, SUVmax values of tumors and reference tissues were significantly higher on the simulated 6.25% ultra-low-dose PET scans as a result of image noise. After the CNN augmentation, the SUVmax values were recovered to values similar to the standard-dose PET. Quantitative measures of the readers' diagnostic confidence demonstrated significantly higher agreement between standard clinical scans and AI-reconstructed PET scans (kappa = 0.942) than 6.25% dose scans (kappa = 0.650). CONCLUSIONS: Our CNN model could generate simulated clinical standard 18F-FDG PET images from ultra-low-dose inputs, while maintaining clinically relevant information in terms of diagnostic accuracy and quantitative SUV measurements.
Entities:
Keywords:
Deep learning; PET denoising; PET/MRI; Pediatric cancer imaging; Whole-body PET reconstruction
Authors: Yan Wang; Luping Zhou; Biting Yu; Lei Wang; Chen Zu; David S Lalush; Weili Lin; Xi Wu; Jiliu Zhou; Dinggang Shen Journal: IEEE Trans Med Imaging Date: 2018-11-29 Impact factor: 10.048
Authors: David J Brenner; Richard Doll; Dudley T Goodhead; Eric J Hall; Charles E Land; John B Little; Jay H Lubin; Dale L Preston; R Julian Preston; Jerome S Puskin; Elaine Ron; Rainer K Sachs; Jonathan M Samet; Richard B Setlow; Marco Zaider Journal: Proc Natl Acad Sci U S A Date: 2003-11-10 Impact factor: 11.205
Authors: Kevin T Chen; Enhao Gong; Fabiola Bezerra de Carvalho Macruz; Junshen Xu; Athanasia Boumis; Mehdi Khalighi; Kathleen L Poston; Sharon J Sha; Michael D Greicius; Elizabeth Mormino; John M Pauly; Shyam Srinivas; Greg Zaharchuk Journal: Radiology Date: 2018-12-11 Impact factor: 29.146
Authors: Christina P W Cox; Daniëlle M E van Assema; Frederik A Verburg; Tessa Brabander; Mark Konijnenberg; Marcel Segbers Journal: EJNMMI Res Date: 2021-07-19 Impact factor: 3.138