A Computerized Detection Scheme For Vertebrae In Computed Tomography Scout Scan

With the rapid development of computers, medical diagnosis gradually transformed fromthe past type of dominated experience to the scientific digital model. Computer-AidedDetection (CAD) has become a research hotspot in the field of medical today. In order toachieve the goal of early detection, diagnosis and treatment, CAD locates the region ofinterest (the lesions area) automatically using the computer. Computer-Aided Detection of thevertebrae is the initialization of the techniques like segmentation and reconstruction. It alsocan automatically mark the scan reference line, which will avoid a lot of manual calibrationwork of the radiographers.Based on the analysis and introduction of the research significance of vertebrae detectionin CT scout scans, the Adaboost learning algorithm is well-studied. Aiming at the problemsthat the existing methods of vertebrae detection requires clear numbers of vertebrae in thepicture, and Haar-like features cannot describe any rotated angles, this paper proposes themulti-angle vertebrae detection algorithm.Firstly, based on Adaboost learning algorithm, the strong classifier is designed. Itconsists of the establishment of the training data set, the calculation of Haar-like features andthe design of the strong classifier. Among them, this paper uses the integral image forcalculating Haar-like features, which greatly increases the speed of operation. This articledesigns the strong classifier using Adaboost learning algorithm. The best Haar-like featuresare collected to distinguish the vertebrae and the non-vertebrae. The corresponding thresholdvalue and the direction are calculated using the method of the minimum classification error.At this point, all the weak classifiers are constructed. A strong classifier is designed bycombining the weak classifiers with its corresponding weights.Secondly, the multi-angle vertebrae detection algorithm is presented. The algorithm cansolve the problems like that vertebrae have different inclined angles in CT scout scans, andHaar-like features cannot describe any rotated angles. Multi-angle vertebrae detectionalgorithm rotates the image according to the inclination angles of vertebrae, so that the vertebrae can be detected in the horizontal position. The vertebrae are detected by traversingimages with the set stride length.Finally, this paper makes the mergence and optimization of the vertebrae detected in thesimilar position of the rotated images.