Meeting Banner
Abstract #0449

Differentiation between Intermingled & Central Type Breast Parenchymal Patterns Using Quantitative 3D Morphology & Texture Analysis on Segmented Dense Tissue

Peter T. Fwu1, Jeon-Hor. Chen1, 2, Siwa Chan3, Dah-Cherng Yeh4, Chin-Kai Chang2, Julian Kao1, Muqing Lin1, Orhan Nalcioglu1, Min-Ying L. Su1

1Center for Functional Onco-Imaging, Department of Radiological Sciences, University of California, Irvne, CA, United States; 2Department of Radiology, China Medical University Hospital, Taichung, Taiwan; 3Department of Radiology, Taichung Veterans General Hospital, Taichung, Taiwan; 4Department of Surgery, Taichung Veterans General Hospital, Taichung, Taiwan

The purpose is to develop a quantitative method to characterize the breast density pattern using 3D morphology and texture analysis for investigating its role in cancer risk prediction. Morphology is characterized by circularity, convexity, and irregularity; texture is characterized by coefficient of variation in signal intensity distribution and the GLCM matrix. The ability of analyzed parameters to differentiate between the central and the intermingled breast density patterns is evaluated by the ROC analysis. The results show that morphology parameters can reach AUC of 0.95, texture parameters have a lower AUC of 0.90, but combining them can yield AUC of 0.98.

Abstract PDF Presentation Video

Keywords

ability abundantly adding addition adipose agents allows alone analyzed appropriate around assign available background better bottom breast calculating cancer carcinogenic central centroid characterize chin china circularity clear coefficient combined computed confined consisted constituted contrast convex convexity curve curves dense density dependent depict depot descriptive determined develop developing development differentiate differentiating differentiation distinguish distinguishable distribution effective energy epithelium evaluated evaluation every facilitate fatty function functional general gland highest histogram homogeneity hospital hull identical improved included incorporated index individual inside inspected intensities intensity intermingled internal investigate involved irregularity link lipid mammary matrix measured mixed morphological morphology never normalized object occurrence orderliness patients pattern patterns percent peripheral peter phys position predicting prediction predictor previously property quantitative radiological radiologist radiology recent reconstructed referred reported representing respectively risk role roughness sciences segmented selected sets shape slice slices soluble sphere statistics still student suggests summarized surface surgery surrounded table texture tissue type variance veterans visually volume yields