A Research On Key Technologies Of On-Demand Medical Film Printing System

With the development of digital medical imaging technology, The PACS system (PictureArchiving and Communication System) has been established in most of hospital. The PACSincludes image acquisition, image browsing, diagnosis, image storage, image processing, andfilm printing management and so on. In most hospitals, medical films are handled and printedby radiology physicians or technician using computer workstation, and then bagged togetherwith DVD and diagnostic report, and finally delivered to the patient. This complicate processis, boring, low efficient and prone to mistakes. Especially, it is more serious-in middle and bighospitals which face up lots of patients and has resource constraints. Therefore, it requires afast and accurate way in medical film printing, packing and delivering, thus leads to anefficient health care service with high quality.This thesis aims to solve the key issues in on-demand film print management system, avirtual print workstation is added between the imaging device and the film printer, to obtainprint package including images and patient information and a database including patientinformation database is defined. The recognition of important information in film isinvestigated in this work.The DCMTK is applied to send and receive messages from image devices and to theprinters. After printing package is detected, and file header is then analyzed. By analyzing thedata flow model in print management and information model referred in DIOCM3.0protocolfor guidance, we design the main process and data flow in detail, especially the systemrequirement analysis, system framework, and database design.Unlike general DICOM image, the header of printing package received by virtual printworkstation does not contain important information, such as patient name, ID, accessionnumber etc., whereas these information is saved in image data as pixels, which is mixed withbackground and other contents. To acquire the above essential information, the patternrecognition scheme is adopted. The samples with different font and size are created, andfeature including projecting, Zernike moment, and wavelet transformation coefficients areextracted, the samples are then trained using radial basis function neural network, To testifythe recognition method, the10-fold cross validation is carried out and quantitative evaluation is performed. Experiments using simulated data and real print package demonstrate thefeasibility and effectiveness of the proposed system architecture design, print data packageanalysis, database design, and automatic text recognition, etc. As the vital parts of on-demandmedical film printing system, the work in this thesis can provide effective solutions andsupport.