2019/06/05 by Hongki Lim, Lim, Hongki, Il Yong Chun +5
Medicine · Physics and Astronomy · #FOS: Computer and information sciences #FOS: Electrical engineering #FOS: Physical sciences #Image and Video Processing (eess.IV) #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Medical Imaging Techniques and Applications #Medical Physics (physics.med-ph) #Nuclear Physics and Applications #Radiation Detection and Scintillator Technologies #Radiomics and Machine Learning in Medical Imaging #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.1906.02327
openalex publication_date 2019/06/05 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Image reconstruction in low-count PET is particularly challenging because\ngammas from natural radioactivity in Lu-based crystals cause high random\nfractions that lower the measurement signal-to-noise-ratio (SNR). In\nmodel-based image reconstruction (MBIR), using more iterations of an\nunregularized method may increase the noise, so incorporating regularization\ninto the image reconstruction is desirable to control the noise. New\nregularization methods based on learned convolutional operators are emerging in\nMBIR. We modify the architecture of an iterative neural network, BCD-Net, for\nPET MBIR, and demonstrate the efficacy of the trained BCD-Net using XCAT\nphantom data that simulates the low true coincidence count-rates with high\nrandom fractions typical for Y-90 PET patient imaging after Y-90 microsphere\nradioembolization. Numerical results show that the proposed BCD-Net\nsignificantly improves CNR and RMSE of the reconstructed images compared to\nMBIR methods using non-trained regularizers, total variation (TV) and non-local\nmeans (NLM). Moreover, BCD-Net successfully generalizes to test data that\ndiffers from the training data. Improvements were also demonstrated for the\nclinically relevant phantom measurement data where we used training and testing\ndatasets having very different activity distributions and count-levels.\n