A Computer-Aided System for Diagnosis of Clustered Microcalcifications in Mammograms

• Main Authors: Chien-Shun Lo, 羅見順
• Other Authors: Pau-Choo Chung
• Format: Others
• Language: en_US
• Published: 2000
• Online Access: http://ndltd.ncl.edu.tw/handle/66349556039025037717

博士 === 國立成功大學 === 電機工程學系 === 88 === Microcalcifications (MCCs) generally present an early sign of breast cancer. In order to assist radiologists in detecting microcalcifications in the early stage of breast cancer, it is highly desirable to develop a reliable computer aided diagnostic system as an assistant. The focus of this dissertation is on the computer-aided system development for diagnosis of clustered microcalcifications in mammograms. Our system is consisted of three subsystems, including the computer-aided design mammography screening system for detection and classification of microcalcifications, the 3-D locating system to find the location of clustered microcalcifications using cranio-caudal and medio-lateral oblique views, and the mammography tele-consultation system. In the first subsystem, the computer-aided design mammography screening system is made up of four modules including Mammogram Preprocessing Module, MCCs Finder Module, MCCs Detection Module, and MCCs Classification Module. In Mammogram Preprocessing Module, breast region is separated from the mammogram by using a region-growing method along with a K-means clustering thresholding method. In MCCs Finder Module, suspicious areas are obtained by screening mammograms using fractal detection method. In MCCs Detection Module, the detection and segmentation of MCCs from the suspicious area is achieved using joint entropy method which is an optimal thresholding method originated from information theory. In MCCs Classification Module, a shape-cognitron neural network is applied to output classifications of MCCs. In the second subsystem, a 3-D locating system is used to display MCCs in a 3-D virtual breast model as a guide for needle biopsy. In this system, three registration features are proposed to register MCCs from two views, and then, 3-D localization of MCCs enables a sequence of coordinate corrections of calcified pixels and display MCCs in the virtual breast model. The three registration features used are gradient code, energy code, and local entropy code originated from statistical theory and information theory. The 3-D virtual breast model is constructed by using virtual reality modeling language (VRML) 2.0 technique, suitable to standard 3D browser. The third subsystem, the mammography tele-consultation system provides a suitable environment for remote mammography consultation, which is developed based on a high speed network, the communication and data security, nice collaboration awareness tools, in conjunction with computer-aided diagnostic system, and linkage of DICOM servers and the PACS system. The key techniques used in this system including distributed network techniques (remote method invoking method, RMI, supposed by JAVA 2), ElGamal algorithm for communication security, collaboration awareness tools including multicursors, meeting lounge, one-to-one talk, E-mail server, file transmission, and so on. In summary, this dissertation provides the odds for diagnosis of clustered microcalcifications by using computer-aided diagnostic system in mammograms. It also provides a prototype and applications of computer-aided diagnostic system for the future five years.