Identification of Breast Lesions by Analyzing Color Images of Ultrasound Elastography

• Main Authors: KAO, WU-YEN, 高于雯
• Other Authors: LEE, WEN-LI
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/g9dd6w

碩士 === 慈濟科技大學 === 放射醫學科學研究所 === 106 === Abstract Traditional B-mode sonography can provide the information of size, shape and echo intensity of a lesion. But it can not show structure properties of the lesion. However, elastography can display and measure the inherent hardness property of the lesion. Literatures report that the benign lesions can be harder or more heterogeneous in texture than the malignants duo to the variety of tissue structures. The fact could result in the difficulty for distinguishing benign and malignant lesions. Diagnostic uncertain like this will make clinicians to appeal pathological biopsy. In this study, a set of elastographic image analysis tools are developed by using an image analysis software. A discrimination method for benign and malignant tumors is also established. It is expected to improve the accuracy for diagnosis of breast tumors without using pathological biopsy on patients. There are two parts in the study. Initially, we acquire elastographic images of the breast phantom, and use image processing software, ImageJ, to perform the color image analysis. The work includes texture, mean and standard deviation analysis. After finding out the relationship between the analytic parameters and hardness of the lesions, the parameters or thresholds is used to distinguish the benign and the malignant lesions. Finally, we apply the method developed in the part one to establish a discrimination method for clinical uses. In the discrimination method established in the phantom experiment, the AUC is 0.89, accuracy is 80%, sensitivity is 100%, and specificity is 67%. 20 lesions are involved in a blind test afterward, its AUC is 0.8. The results in clinical image analysis, Kappa value is 0.8 (p = 0.000). It indicates that the use of predictive values to distinguish between benign and malignant tumors is significantly correlated and consistent. We develop a fast and convenient image analysis method to distinguish benign and malignant tumors. It can be a useful method for clinical diagnosis of breast tumors, and reduce unnecessary pathological biopsy on patients.