Abstract
1670
Objectives: The objective in this work is to use image inpainting to model backgrounds in nuclear medicine.Examples are drawn from nuclear cardiology where we apply this technique to sinogram space.
Methods: Digital inpainting is a mechanism whose objective is to fill missing parts of an image using its own information in an automatic way. The area with missing data is called the inpainting region and should be filled such that both structure(edges) and texture information remain coherent corresponding to the background modelling zone. We apply wavelets to the inpainting problem. The wavelet transform(WT) consists of the application a function called the mother wavelet to a image which allows to identify its frequency components and their spatial location. The transform performs convolutions of the input image with low pass and directional high-pass filters and iteratively downsampling the output at each step. The method takes an inage I and a user defined inpainting mask corresponding to the required background interpolation region and decomposes both images using a decimated WT. Wavelet coefficients both low pas and detail coefficients are the propagated into the inpainting region, and the inverse WT is applied to obtain the final reconstructed background interpolated image. At each iteration , a block with varying size is selected as the filling target based on its geometric aspects and the energy of the wavelet coefficients in neighbouring regions.
Results: This approach is demonstrated in myocardial SPECT where a masking region is defined in sinogram space bordering the myocardium . Inpainting is used to model and interpolate the underlying activity including subdiaphramatic activity , a common cause of artefact.
Conclusions: Inpainting provides a novel and accurate approach to background interpolation in nuclear medicine.
- Society of Nuclear Medicine, Inc.