Multi-scale processing of tomographic images using dyadic wavelet expansions

In nuclear medicine, clinical radiological data often have limited image quality due to the safety requirements in dose level for X-ray and radionuclide modalities. In this dissertation, new techniques of multi-scale de-noising and enhancement were investigated to improve image quality of tomographic data while reducing dose admissions.; A multi-scale adaptive histogram equalization (MARE) technique was developed to achieve simultaneous enhancement of multiple image features across wide dynamic ranges in a single image. An evaluation study with 109 clinical chest CT cases was carried out to validate the efficiency of MARE as a pre-processing tool.; We generalized the conventional multi-scale thresholding scheme such that each multi-scale sub-band is processed with a distinct thresholding operator. Such a paradigm granted more flexibility to the process of designing an effective thresholding rule for de-noising and enhancement. In addition, a cross-scale regularization process was designed to effectively recover detail signal features within multi-scale sub-bands.; The effectiveness of multi-scale adaptive thresholding and cross-scale regularization were systematically evaluated using both phantom and clinical datasets. For Positron Emission Tomography (PET) and Single Photon Emission Computed Tomography (SPECT) imaging, it showed consistent improvement in image quality when compared to existing techniques in a clinical comparison study using 30 PET brain data. The proposed de-noising techniques were also utilized as an optimization criterion for tomographic reconstruction using Filtered Back-Projection (FBP). Three dimensional rotational X-ray imaging also benefit from multi-scale adaptive thresholding. In a comparison study using 20 clinical spine data and 20 clinical angiography data, both quantitative measurement of image quality and qualitative comparison with three-dimensional volume rendering showed consistently superior noise-removal results when compared to existing pre-processing techniques applied in current commercial clinical systems.; This dissertation also reports on research efforts on image segmentation. A hybrid segmentation paradigm was demonstrated and validated by a clinical study of quantifying adipose tissue using whole body MRI (magnetic resonance imaging) scans.; Potential clinical application of the developed multi-scale de-noising techniques was shown in many aspects, including the reduction of scan time (PET), X-ray dose level (3DRX), and interpretation complexity (CT). We also showed that effective de-noising improves further image analysis including segmentation and quantification by reducing variability.