Study on Computer-Aided BI-RADS iagnosis and Classification for Breast Ultrasound

• Main Authors: Wei-Chih Shen, 沈偉誌
• Other Authors: Ruey-Feng Chang
• Format: Others
• Language: en_US
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/72480914604779106369

博士 === 國立中正大學 === 資訊工程所 === 96 === With the rapid development of computer applications, many researchers attempt to developed a computer-aided diagnosis (CAD) system to help the radiologists for differentiating the benign from malignant masses. In the clinical diagnosis, radiologists, however, classify the masses into assessment categories probably benign, suspicious abnormality, and highly suggestive of malignancy according to the likelihood of malignancy which is evaluated by the descriptions of mass sonographic characteristics defined in BI-RADS. In this thesis, the clinical diagnosis process of radiologists was simulated as the computer application system. Firstly, the descriptive items of sonographic characteristics defined in BI-RADS were mathematically quantified and the diagnostic performance of proposed computerized BI-RADS features for CAD system were measured. Then, a computer-aided classification (CAC) system which could be used to diagnose the cases acquired form different ultrasound systems was proposed. Compared with the currently developed CAD systems, the indeterminate cases were classified to category suspicious abnormality for further proving in the proposed CAC system. Besides, the influences of spatial compound imaging technique for the diagnostic performances of CAD system was assessed. In the proposed computerized BI-RADS features, the contributions of margin characteristics, lobulate and angular, was the most significant for the diagnostic performances of CAD system and that of CAC system. The interobserver agreement between the radiologists and the CAC system was indicated as substantial agreement. Compared with the radiologists, the specificity was improved by 8.85% in the proposed CAC system. Compared with the conventional CAD system, the sensitivity was improved by 10.5% in the proposed CAC system. And, we concluded that the spatial compound imaging technique not implies the improving of diagnostic performance for CAD system even though the quality of ultrasound images were improved.