Preclinical CT

Within the last decade micro-computed tomography (micro-CT) advanced to an essential tool in preclinical research and is thus also referred to as preclinical-CT. Early applications aimed at determining the internal structure of biological specimen in an ex-vivo manner, e.g. bone explants, with a spatial resolution of up to 1 µm. Recent developments in flat detector technology and a consequent reduction in acquisition time enabled the imaging of small animals, usually mice and rats, in-vivo with resolutions as low as a few micrometers. Current in-vivo applications include, but are by far not restricted to, phenotyping, imaging of tumors and metastasis, including the vasculature and neovascularization, assessment of pulmonary and cardiac function, visualization of fracture healing and vascular exploration (see figure 1). A special focus of our research group is respiratory and cardiac imaging in mice and rats. This is motivated by a variety of reasons such as the fact that a multitude of mouse models of cardiac diseases has been established in preclinical research. Other possible applications include tumor tracking during respiration for the assessment of lung cancer progression or of pulmonary diseases in general.

The rapid heart and respiratory rates of free breathing mice and rats of up to 600 bpm (beats per minute) and 300 rpm (respirations per minute) usually prohibit in-vivo imaging studies even if gating methods are applied. Often, the animal needs to be sacrificed and examined ex-vivo. Furthermore, to allow for longitudinal studies, the radiation dose needs to be kept at a minimum to prevent any metabolic interferences. To overcome these issues our research group proposed several novel reconstruction paradigms. In a standard in-vivo micro-CT scan the respiratory motion and ECG signals are extracted from the rawdata and used to select the projections appropriate for the given heart and respiratory phases. Conventionally, double-gating is performed to obtain images without motion artifacts. Those phase-correlated (PC) images, however, suffer from a high noise level as only a small number of the total projections match the desired motion phases (respiratory and cardiac phases need to be matched simultaneously). We, for example, designed an iterative approach using a priori information from all projections combined with 5D (spatial, cardiac and respiratory) anisotropic edge-preserving filtering to overcome this drawback and significantly improve the image quality [1]. Using this low-dose phase-correlated (LDPC) reconstruction the voxel noise is typically reduced by a factor of up to six and artifacts are almost removed. The radiation dose of our standard protocol is about 500 mGy. Reducing the number of projections available for image reconstruction and hence the administered radiation dose illustrates that one can achieve comparable image quality with only 200 mGy of dose when using LDPC. Compared to other publications that apply 1840 to 2400 mGy dose and use PC reconstruction (similar spatial resolution and image noise), our LDPC approach therefore achieves a more than ten-fold dose usage improvement and allows for longitudinal studies of cardiac and pulmonary diseases in small animals. Similar results were achieved using our high-dimensional total variation (HDTV) minimization algorithm [2]. This method, derived within the compressed sensing framework, achieves the reconstruction of a volume by optimizing the rawdata fidelity and the five-dimensional gradient (spatial, cardiac and respiratory) between adjacent cardiac and respiratory phases. The newest flavor of algorithms estimates the motion vector fields between adjacent cardiac and respiratory phases [3]. These so called motion compensation (MoCo) methods allow for a deformation of the volume achieved using the data for each cardiac and respiratory phase to a given reference phase. Hence, all reconstructed volumes can be superimposed allowing for an optimal usage of administered radiation dose, as all projections are used for the reconstruction of each volume. The resulting volumes exhibit highest image quality as the noise level of a given motion phase is equivalent to the noise level of a Feldkamp reconstruction using all available data (see figure 2).