Purpose: 4D (3D plus time) CT data, such as cardiac or perfusion images, are frequently affected by body motion. Additionally, these data are normally very noisy due to the use of lower scanner output settings in order to reduce patient dose. State‐of‐the‐art denoising filters can achieve efficient noise reduction of CT images, but fail to handle data containing significant body motion. The purpose of this work was to develop a filter that can overcome this limitation. Methods: A modified Partial Time Vector NonLocal Means filter (PTV‐NLM) recovers image from the original noisy data by representing a pixel using the weighted average of its neighbors in both the spatial and temporal domain. The weighting factor is determined by calculating the summed square difference of portions of the temporal profile of each pixel with its neighboring pixels. This approach does not require motion estimation, can automatically adapt to local motion and efficiently utilizes the redundant information in the images. Results: The denoised images showed better image quality than the original images, making anatomical structures easier to identify in the denoised images. After denoising, the contrast to noise ratio of simulated low‐dose perfusion data (using 25% of the full dose) was comparable to that of images acquired using the full dose. Alternatively, for applications such as cardiac CT, PTV‐NLM can be used with the full dose data to decrease image noise without introducing spatial blurring. Conclusion: In 4D data sets, the PTV‐NLM filter can maintain image quality at dramatically decreased dose levels, or provide better image quality at the same dose levels. It can be used to either reduce patient dose or improve image quality without causing blurring of moving anatomy.
ASJC Scopus subject areas
- Radiology Nuclear Medicine and imaging