A Study Of The Cerebral CT Image CAD Algorithm Based On Texture Feature

With the rapid development of science and technology, new and modern equipments such as Multi-slice Spiral Computed Tomography have become the leading instruments for medical imaging. These equipments produce volume data of high quality and large quantity. While volume data provides more and more detailed and accurate diagnostic information for the radiologist, it also increases significantly the workload of radiologists. Overload would exacerbate the radiologist's fatigue and increase the possibility of overlooking and mistaking in the course of film reading. Instead of making diagnosis easier,new equipment puts radiologists into a dilemma and its advantages have not been exploited. As these increasingly advanced equipments are rapidly and widely put into use, the incompatibility between modern equipments and traditional human film reading would become even more evident and more acute. To keep pace with the development of modern medical imaging technology, research directing to realize Computer Aided Diagnosis or Computer Intelligent Diagnosis has become the major challenge and general trend in the field of medical imaging processing and analysis.The first and biggest obstacle to clear on the way to computer intelligent diagnosis is how to detect the lesion on the medical image. In order to successfully detect pathology on medical image, many of the image processing techniques, such as registration, segmentation and construction of digital probabilistic atlas have to be studied. With the direction of future research on Computer Intelligent Diagnosis, with the goal of pathology automatic detection based on the texture information on the cerebral Computed Tomography, we heve made a in-depth study on the method of pathology automatic detection and its related technique by combining domain knowledge of medical image, and we have put forward a few of innovative algorithm. The major contributions of the dissertation are as follows: 1. A method which carries out computer-automatic diagnosis according as the texture feature of the cerebral CT image was put forward, and a kind of Texture Layers-Analysis Vector based on tree-structured wavelet transform which characterizes the texture feature of the cerebral CT image was studied and constructed in this paper. After researching the computer-automatic diagnosis algorithms, the texture information of image was attached importance to, which was ignored in the other algorithms. Because the texture information characterizes the relationship between the pixels in the image, and the wavelet transform offers the analysis and the expressing of the texture feature a kind of exact and uniform explanation, the Texture Layers-Analysis Vector is able to characterize the texture information of the cerebral CT image exactly. Experiment results showed that the texture layers-analysis vector is robust, unique in the image, and it is invariable after shifting and rotating in a range. The research of the Texture Layers-Analysis Vector facilitates the work in this paper later.2. An algorithm of automatic detection to corresponding points based on the texture layers-analysis vector was proposed in cerebral CT images. NRR is the key step to pathology detection, therefore a variety of algorithms of NRR was first reviewed and introduced in this paper. A algorithm of automatic detection to corresponding points based on the texture layers-analysis vector was put forward later because the existing NRR algorithm is not automatic completely. In this algorithm the point which has similar texture information is pick out as the corresponding point. Experiment results showed that the correct rate of this algorithm is satisfying.3. The digital probabilistic atlas based on texture layers-analysis texture in 2D cerebral CT image was constructed automatically. The digital probabilistic atlases were reviewed in this paper. We proposed the digital probabilistic atlas algorithm based on the texture layers-analysis vector according to the results before because the other atlases have a shortage of characterizing the normal feature of cerebral CT image. In this paper, the digital probabilistic atlas based on the texture layers-analysis vector was constructed automatically. 4. Using the digital probabilistic atlas based on the texture layers-analysis vector, we detected calcify, cerebral hemorrhage, etc automatically with computer. Because most representation of the pathology is the texture changes, according to comparing the difference between the texture layers-analysis vectors which characterizes the texture information of the cerebral CT image, a variety of pathological changes such as calcify, cerebral hemorrhage was detected automatically.