Research On Detection Approach For Pulmonary Nodules Based On CT Images

According to the clinical research, lung cancer occupies the first position in bothmorbidity and mortality, and has become one of the deadly diseases in the world. Lungcancer shows solitary pulmonary nodule in the early stage. It will reduce mortality greatlythat patients can detect pulmonary nodules and be cured in the early stage. Therefore, inorder to decrease mortality, it is crucial to improve the accuracy of the early detection.Computed Tomography is considered as one of the best methods to diagnose thepulmonary nodules. However, a large number of CT images also increase the work for theradiologists to read them, leading to increasing omission diagnosis rate and misdiagnosisrate. Along with the continued improvement of the digital image processing technology,medical imaging technology and pattern recognition technology, computer-aided detectiontechnology can develop rapidly. Making use of computer-aided detection for lung nodulescan not only provide reference information for the radiologist, but also improve thedetection efficiency and accuracy. Therefore, research on computer-aided detection will beof great importance and value.According to the subject of detection for solitary pulmonary nodules based on CTimages, this paper mainly analyses the shortages of the existing methods, and proposesresearch targets and solutions of the problems. At last, the paper achieves automaticdetection for solitary pulmonary nodules.Firstly, the algorithm implements wiener filter processing on original CT images,which can get better denoising result. Secondly, the paper adopts adaptive iterationthreshold method twice to implement initial segmentation of the pulmonary parenchyma,and achieves complete pulmonary parenchyma automatically by using region-mark andcontour extract method. Furthermore, the experiment makes use of adaptive thresholdmethod to locate candidate nodules by combing the histogram analysis with the feature ofsolitary pulmonary nodules on CT images, and then achieves feature extraction for ROIs.Finally, the algorithm employs SVM to construct classifier to classify and recognize true solitary pulmonary nodules, and then labels the nodules’ positions on original CT images.According to the standards of performance evaluation, the accuracy of the detection canachieve93.407%, as well as the specificity can achieve98.485%.Experimental resultsshow that our algorithm is feasible and can detect solitary pulmonary nodules effectively.