Abstract | ||
---|---|---|
We propose a method based on the generative adversarial networks to obtain diagnostic-quality volumetric images (magnetic resonance imaging (MRI) and computed tomography (CT)) from fast data acquisition protocols and lower radiation dosage respectively. These types of acquisition protocols are beneficial to patients, but often inadequate for diagnostic purposes due to the loss in image quality. This work extends the approach presented in [1] to volumetric medical imaging by using perceptual and clinically meaningful loss function. The proposed loss function consists of adversarial and perceptual loss components. The adversarial loss is trained to differentiate between the estimated and the diagnostic quality image while the perceptual loss incorporates perceptual similarity instead of pixel-wise similitude. The adversarial architecture with the perceptual loss was separately trained on 40 CT and 40 MR images, respectively and subsequently used to enhance low resolution scans to obtain diagnostic quality image data. The proposed framework produced an average peak signal-to-noise ratio (PSNR) of 32.09 ± 0.24 dB and 33.76 ± 0.28 dB for CT and MR images respectively in our experiments. The quality obtained is greater than the minimum discernible PSNR threshold of 31.7 dB of the human visual system, indicating high quality for diagnosis [2]. |
Year | DOI | Venue |
---|---|---|
2018 | 10.1109/ISBI.2018.8363591 | 2018 IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018) |
Keywords | Field | DocType |
Image enhancement,Radiation dosage,Fast image acquisition,Generative adversarial network | Computer vision,Similitude,Pattern recognition,Computer science,Medical imaging,Human visual system model,Data acquisition,Image quality,Artificial intelligence,Computed tomography,Magnetic resonance imaging,Perceptual similarity | Conference |
ISSN | ISBN | Citations |
1945-7928 | 978-1-5386-3637-4 | 0 |
PageRank | References | Authors |
0.34 | 0 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Awais Mansoor | 1 | 68 | 12.49 |
Teerit Vongkovit | 2 | 0 | 0.34 |
Marius George Linguraru | 3 | 362 | 48.94 |