Dose reduction and image enhancement in preclinical mouse imaging using deep learning
Author and co-authors: Florence Muller, Jens Maebe, Boris Vervenne, Joel Karp, Christian Vanhove, Stefaan Vandenberghe
Preclinical PET and CT imaging provide a powerful toolset to non-invasively acquire functional and anatomical images of laboratory animals, yet both modalities involve ionizing radiation. While delivered dose levels are normally non-lethal to the animal, they can be substantial enough to impact experimental outcomes of animal models, especially in longitudinal follow-up studies. It is thus important to aim for dose reduction, but lowering the radiation dose inherently introduces noise in the images, and reduced image quality negatively impacts diagnostic performance. Various denoising techniques already exist, but deep learning (DL) methods have become increasingly popular for image quality enhancement.
In this webinar, ir. Florence Muller (Ghent University – University of Pennsylvania) will present two recent studies that aimed to investigate the use of convolutional neural networks (CNN) to denoise low dose micro-CT and micro-PET images. Florence will explain how she developed and evaluated an image-to-image CNN framework to predict higher quality images from noisier images acquired at lower radiation doses for both modalities.