Design And Implementation Of An Intellectual Detection Algorithm Of Standard Hand Position In X-ray Radiography

Since X-ray was discovered more than100years ago, It has been widely used todevelop various analog and digital medical imaging devices, from screen filmimaging system to fashion digital radiography. Various X-ray-based medical imagingdevices are widely used in modern hospitals. The main factor that influences imagequality is the accuracy of radiography positions in clinical application. An algorithmis proposed, which can help doctors to judge the right radiography position, thuseffectively reduce the wrong shot and protect the patient from the radiation dose fromreplicate shots. Meanwhile the algorithm brings improvement to the image quality andadvance the efficiency of X-ray examination.In the thesis, computer vision technology is applied to intellectually recognize thesix most common standard positions (hand AP, hand oblique PA, hand oblique AP,thumb AP, thumb lateral and single finger lateral) used in present clinical X-rayradiography. According to the X-ray radiography position standard released by WorldHealth Organization, An algorithm which is used to judge the standard X-rayradiography positions is developed by independent research. It can recognize sixkinds of most common hand position including hand AP, hand oblique PA, handoblique AP, thumb AP, thumb lateral and single finger lateral. To reduce the influenceof illumination variations and complex background, the hand contour extraction partof the algorithm was improved. Firstly, codebook was adopted to model thebackground for extracting the foreground; then the foreground and background’ sskeleton topology, of which watershed calculation took advantage to obtain the profileof the hand, were extracted by thinning algorithm. We established a platform onwhich t the algorithm can be tested and the experimental data of radiography positionscan be collected. Finally the software of collecting Web video data was programmed.Instructed by a medical expert, we collected100cases of experimental data by theplatform to evaluate the validity and accuracy of the recognition algorithm.The experiment results showed that the standard and misplacing positions ofthumb AP, thumb lateral, single finger lateral and hand AP could be identified by thealgorithm with an accurate rate of over90percent. However, because of lacking thedata of depth of field, the accuracy rate of hand oblique PA and hand oblique AP’s recognition is only about80percent. In a word, the algorithm basically realizedradiography position recognition. At the same time, the design of the algorithm takesbackward compatibility into account, which is very convenient for further extendingand updating to satisfy the user’s requests and fix the weakness of the algorithm.