/NM-Img-Denoising-DIP-Keras

Primary LanguagePythonMIT LicenseMIT

Denoise Nuclear Medicine images with the Deep Image Prior

Deep Image Prior (DIP) is discribed in:

Ulyanov, Dmitry, et al. "Deep Image Prior." The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018, pp. 9446-9454.

This is an application of DIP in image denoising in reconstructed 2D PET images.

Sample Results:

For simplest testing, input noisy image is a 2D slice of a 3D PET image corrupted by the additive Gaussian noise (Var=0.001, 0.005, and 0.01).

Variance = 0.001:

Variance = 0.005:

Variance = 0.01:

About Myself