Thivya K S, Sakthivel P, Venkata Sai P M
Department of Electronics and Communication Engineering, Anna University, Chennai, India.
Department of Radiology & Imaging Sciences, Sri Ramachandra Medical Centre, Chennai, India.
Technol Health Care. 2016;24(1):21-9. doi: 10.3233/THC-151042.
Breast cancer is the second threatening tumor among the women. The effective way of reducing breast cancer is its early detection which helps to improve the diagnosing process. Digital mammography plays a significant role in mammogram screening at earlier stage of breast carcinoma. Even though, it is very difficult to find accurate abnormality in prevalent screening by radiologists. But the possibility of precise breast cancer screening is encouraged by predicting the accurate type of abnormality through Computer Aided Diagnosis (CAD) systems. The two most important indicators of breast malignancy are microcalcifications and masses. In this study, framelet transform, a multiresolutional analysis is investigated for the classification of the above mentioned two indicators. The statistical and co-occurrence features are extracted from the framelet decomposed mammograms with different resolution levels and support vector machine is employed for classification with k-fold cross validation. This system achieves 94.82% and 100% accuracy in normal/abnormal classification (stage I) and benign/malignant classification (stage II) of mass classification system and 98.57% and 100% for microcalcification system when using the MIAS database.
乳腺癌是女性中第二大致命肿瘤。降低乳腺癌风险的有效方法是早期检测，这有助于改善诊断过程。数字乳腺钼靶摄影在乳腺癌早期的钼靶筛查中发挥着重要作用。然而，放射科医生在大规模筛查中很难发现准确的异常情况。但是，通过计算机辅助诊断（CAD）系统预测准确的异常类型，有望实现精确的乳腺癌筛查。乳腺恶性肿瘤的两个最重要指标是微钙化和肿块。在本研究中，研究了一种多分辨率分析方法——小框架变换，用于对上述两个指标进行分类。从不同分辨率水平的小框架分解乳腺钼靶图像中提取统计特征和共生特征，并采用支持向量机进行k折交叉验证分类。使用MIAS数据库时，该系统在肿块分类系统的正常/异常分类（I期）和良性/恶性分类（II期）中分别达到了94.82%和100%的准确率，在微钙化系统中分别达到了98.57%和100%的准确率。
