Quality Enhancement and Segmentation for Biomedical Images

This dissertation, "Quality Enhancement and Segmentation for Biomedical Images" by Hongmin, Cai, 蔡宏民, was obtained from The University of Hong Kong (Pokfulam, Hong Kong) and is being sold pursuant to Creative Commons: Attribution 3.0 Hong Kong License. The content of this dissertation has not been altered in any way. We have altered the formatting in order to facilitate the ease of printing and reading of the dissertation. All rights not granted by the above license are retained by the author. Abstract: Abstract of thesis entitled QUALITY ENHANCEMENT AND SEGMENTATION FOR BIOMEDICAL IMAGES submitted by CAI Hong-Min for the degree of Doctor of Philosophy at The University of Hong Kong in October 2007 The main task of this study was to reconstruct the 3D image of a mouse axon from its 2D cross-section images. New mathematical methods were designed to achievethis. The3Dreconstructioninvolvesthreesteps: preprocessing, segment- ingandtrackingoftheimages. Foreachstep, commonlyusedmethodshavebeen appliedbuttheresultswereunsatisfactory, henceanewmethodforeachstepwas designed. Thenewlydesignedmethodsnotonlyworkwellinthemiceaxonrecon- struction, butalsocanbeappliedtootherimagesofsimilartasks. Sincedi(R)usion methods are widely used for biomedical image preprocessing and commonly used methods were found to have shortcomings when applied to mice axon images, a chapter of the thesis was to review these methods and their shortcomings. After that, a new di(R)usion scheme for preprocessing, a snake based method for semi- automatic segmentation and a framework for fully automatic segmentation and tracking were described in individual chapter. Since all our new methods were motivated by the shortcomings of those commonly used methods when applied to mice axon images, detailed explanations were provided. For the preprocessing step, denoisingandfeatureenhancingarethemaingoals. Currentmethodseither over-smoothen or leave clusters that could cause false edges in images. To over- come these, the new method has an additional "gradient vector ow" term that balancesbetweenover-smoothingandstructuraledgespreservation. Experimentsshowedthatitworkedverywellonawidevarietyofimageswithdi(R)erentnatures, especially on axonal images. The second step in the reconstruction is segmen- tation. Again, currently used methods su(R)ered from poor quality of the axon images, especially when both strong and weak boundaries are present, and gave misleading segmentation results. The new method has an extra repulsive feature that can avoid overwhelming of the strong boundaries over the weak ones. It was further rened by adding a shape constraint and produced very good segmenta- tion results. The last step of the reconstruction was to piece the 2D cross-section images into a 3D image with each axon clearly identied and colored di(R)erently from each others. It was done by adaptive mathematical morphological opera- tions after the di(R)usion preprocessing. Since splittings and mergings of axons can occur among images, special devices such as mean shifts were used to handle them. Finally, all the reconstruction steps were assembled into a self-contained automatic framework that can be used to track objects with topological changes. DOI: 10.5353/th_b3938013 Subjects: Diagnostic imaging Image reconstruction Three-dimensional imaging