علي نادري پاريزي

Breast cancer remains a major global health challenge and is one of the most common cancers affecting women. With its increasing incidence rate and profound impact on patient well-being, early and accurate diagnosis is crucial for improving patient outcomes. In this context, medical imaging technologies have emerged as essential tools, providing physicians with non-invasive methods for detecting and assessing breast abnormalities. Among these, Automated Breast Ultrasound (ABUS) has gained prominence due to its ability to provide a comprehensive eva‎luation of breast tissue, serving as a complement to traditional mammography. With the growing emphasis on precision medicine, Computer-Aided Diagnosis (CAD) systems have become valuable tools in breast cancer detection. In this study, various classification methods—including deep learning-based approaches and traditional feature extraction techniques—were tested on a new public dataset, TDSC, and the distinguishing characteristics of tumors were examined. Research revealed that focusing on texture irregularities, protrusions, and indentations along tumor boundaries is the most critical feature for classifying tumors as benign or malignant. The highest accuracy (70%) and an ROC AUC of 0.83 were achieved when 3D tumor boundary features were applied to a Support Vector Machine (SVM) classifier, which is competitive with current research. Given that similar studies indicate that classifying breast tumors in 3D ultrasound images remains an unsolved problem, future research should focus on tumor boundary characteristics. Additionally, more publicly available datasets are needed to achieve more reliable and satisfactory results.

رده‌بندي , اولتراسوند 3بعدي , پردازش تصوير پزشكي , يادگيري عميق

classification , 3d ultrasound , medical image processing , deep learning

Ali Naderi Parizi

Mohsen Soryani