Vessel Detection And Quantization Technique Of Three-dimensional Medical Images

Currently, the vessel diseases are the most harmful diseases with high morbidity,mortality and disability rate. The diagnosis of vessel diseases is heavily dependent onthree dimensional CT images. However, the information embedded in the three dimen-sional CT images is too huge to humanbeings. This gives a call for Computer-aideddiagnosis system. With the developing of the research on vessel diseases, as well as theapplication promoting, vessel detecting algorithm faces a growing number of targets withspecial circumstance, such as vessels with incomplete features or uncertain characters.The vessels with such circumstances appear in the pathology regions and give a rise forclinical needs with high integrity and high accuracy detecting. Due to the incomplete anduncertain features, the performance of current algorithm dose not meet the clinical needs.The main purpose of the present paper is facing the vessels with incomplete anduncertain features to design an algorithm, whose results are complete and with high ac-curacy. Starting with the property of vessels, following with the integration of imageprocessing, artificial intelligence and pattern recognition technology, the algorithm aimsto analyze and solve biological problems by mean of numerical methods. Specific contentof this paper is as follows:(1) Aimed at the incomplete detecting results with vessels of low contrast, the presentpaper proposed an algorithm to fix the incomplete centerlines with optimization basedon smoothness detecting. The traditional algorithms can not result in complete center-line, due to the uncertain features caused by low contrast. Overcoming the incompleteproblem, the present paper described the uncertainty of global repair of the incompleteproblem based on smoothness detecting, and gave global optimal treatment to repair theincomplete centerlines by global optimization algorithm. Additionally, the global opti-mization algorithm got more accurate results by suppressing noise.(2) Aimed at small vessels with low signal to noise ratio, low contrast, and nearby tis-sue disturbance, the present paper proposed a constrained B-spline active contour model.The traditional algorithms can not detect small vessel accurately, due to the low contrast.The small vessels come with low signal to noise ratio and nearby tissue disturbance whichpropose great challenges for detecting small vessels accurately. To overcome the abovelimitations, the present paper proposed a constrained B-spline active contour model. Theproposed model formulates energy functions base on geometric features of vessels, andintegrates the evolution of active contour via Euler-Lagrange optimization method. Theconstrained active model can detect small vessels in special circumstance. In addition, theconstrained B-spline active contour model can tiny vessels due to the combination withB-spline which accelerates the convergence.(3) Aimed at abnormal vessels with high undetected rate and high error rate, thepresent paper proposed a sparse representation algorithm base on Patch. The abnormalvessels are the greatest challenges in detecting vessels which show up in pathological re-gions. Due to the complex environment of pathology regions, the abnormal vessels losttheir features. In such circumstance, the undetected rate and error rate of the traditionalalgorithms are too high to meet the clinical needs. To overcome the above limitations,the present paper proposed an algorithm based on Patch implicit features which avoidsfallings by using explicit features. Facing the high time complexity of Patch based al-gorithm, the present paper proposed a sparse representation algorithm to accelerate theproposed algorithm, and make it with allowing easy application.(4) The present paper compared the detecting results by analyzing quantitatively.The validity and accuracy are the key of an algorithm to its clinical application, whilequantization is the basis of judging validity and accuracy of an algorithm. From multi-ple angles with multiple quantization criteria, The present paper quantified the algorithmresults and analyzes the performance of the prosed algorithms. To understand the advan-tages of the proposed algorithms intuitively, he present paper compared the performanceof a set of state-of-art algorithms. The comparison of the quantization results verified thevalidity, accuracy and practicality of the proposed algorithms.