LOW DOSE CBCT DENOISING USING A 3D U-NET

Cone-beam computed tomography (CBCT) is an essential tool to medical practitioners as it allows obtaining a three-dimensional (3D) representation of a scanned body part. These representations are often used in image-guided procedures. However, acquiring these CBCT scans requires subjecting the patient with a sufficient radiation dose in order to obtain a high quality 3D image. Although relatively low, the radiation dose associated with CBCT can increase the lifetime risk of a secondary malignancy. By reducing the radiation dose, there is greater noise within the images, reducing their clinical efficacy. Therefore, there has been numerous works focused on low dose CBCT that use either conventional image processing-based or deep learning-based denoising methods. While deep learning-based methods are generally superior, they typically consider the 3D CBCT volume as separate 2D images. In this work, we develop a 3D deep learning-based framework to denoise low dose CBCT scans. Our results show that we are able to remove a significant portion of the noise from the input volume.