Font Size: a A A

Research On ROI Extraction For Lung CT Images

Posted on:2012-02-18Degree:MasterType:Thesis
Country:ChinaCandidate:A L YangFull Text:PDF
GTID:2218330362953642Subject:Biomedical engineering
Abstract/Summary:PDF Full Text Request
The rapid growth in the number of medical images requires a strong development of Content Based Medical Image Retrieval technology. The effective Content Based Medical Image Retrieval method plays an important supporting role in the advance of medical field. In order to solve this problem, first of all, a problem we faced is that how to extract the regions of interest in medical images (Region of Interest, ROI) fast and accurately. Medical image information mainly concentrated in the area of the lesions, the doctor medical diagnosis, treatment, case study and other operates of patients are based on the analysis of the lesion areas.CT image is one type of medical imaging which is widely used in the medical field. With the changes in the human living environment and lifestyle, pulmonary disease has become one of the most severe malignant tumors which seriously impact on human health, especially the lung cancer is more harmful to human life. Standing the perspective of human health, based on the lung CT medical images, we do the exploration and research about region of interest extraction. The technology that region of interest extraction has profound meaning for the medical image retrieval and pattern recognition.This paper presents a fusion of one-dimensional Otsu method and mathematical morphology segmentation method of lung parenchyma, this method makes use of lung CT image's gray-scale information, and this method can segments the lung parenchyma effectively. Based on the segmentation of lung parenchyma, we use a fusion of two-dimensional Otsu method and mathematical morphology to extract the regions of interest, and then after analysis of the main characteristics of visual lesion features based on region of interest, we judge and remove the interference noise regions. Lastly, we extract the HOG characteristics of different types of signs, and then use the MMP classifier to train the sign models which can be used to classify the ROI candidates, HOG features can excellently represent lung sign characteristics, while the MMP classifier can learn sign model well.In this paper, all experimentations were based on the real cases of lung CT images. Our fusion method of 1D OTSU and mathematical morphology is effective for lung segmentation, which accuracy rate of Case and Image can reach 92.14% and 89.71% respectively; 2D OTSU and mathematical morphology combined extraction method of region of interest can extract most of the lesions in the lung regions; and based on HOG features of ROIs, our classification method can remove most of the invalid ROIs, the sensitivity and specificity can reach 53.85% and 88.68% respectively, the overall accuracy rate is 88.67%, and the accuracy of our method for calcification lung CT is 36.84%。...
Keywords/Search Tags:CT of Lung, lung segmentation, Region of Interest extraction, Feature extraction
PDF Full Text Request
Related items