Heang-Ping Chan PhD, Presenter: Nothing to Disclose

Yi-Ta Wu PhD, Abstract Co-Author: Nothing to Disclose

Berkman Sahiner PhD, Abstract Co-Author: Nothing to Disclose

Yiheng Zhang PhD, Abstract Co-Author: Nothing to Disclose

Richard H. Moore MD, PhD, Abstract Co-Author: Research support, General Electric Company

Daniel B. Kopans MD, Abstract Co-Author: Research support, General Electric Company

Mark Alan Helvie MD, Abstract Co-Author: Institutional grant, General Electric Company

Lubomir M. Hadjiiski PhD, Abstract Co-Author: Nothing to Disclose

Ted Win Way, Abstract Co-Author: Nothing to Disclose

et al, Abstract Co-Author: Nothing to Disclose

et al, Abstract Co-Author: Nothing to Disclose

To analyze characteristic features of masses on DBT mammograms.

DBT cases were obtained with a GE Gen1 prototype system at the Mass General Hospital with IRB approval. The DBT system acquired 11 projections over a 50-deg arc. A simultaneous algebraic reconstruction technique was used to reconstruct the DBT at 1-mm slice spacing. The mass on each slice was segmented by an active contour model initialized with adaptive k-means clustering. A spiculation likelihood map was generated by analysis of the gradient directions around the mass margin and spiculation measures were extracted from the map. The rubber band straightening transform (RBST) was applied to a band of pixels around the segmented mass boundary. The RBST image was enhanced by Sobel filtering from which run-length statistics (RLS) texture features were extracted. Morphological features including those from the normalized radial length (NRL) were designed to describe the mass shape. The corresponding features were averaged from a number of slices centered at the slice where the mass was best visualized. A data set of 60 cases with 30 malignant and 33 benign masses was used in this preliminary study. Linear discriminant analysis with stepwise feature selection identified useful features and designed their weights using the training set in a 3-fold cross validation scheme. Several 3-fold random groupings were evaluated and the results were averaged.

An average of 3 or 4 features was selected, depending on the number of slices used. Two spiculation measures and one RLS texture feature were found to be most effective for differentiation of malignant and benign masses. An NRL mass shape feature was also frequently selected. The test Az ranged from 0.90±0.04 to 0.93±0.03 as the number of slices for feature averaging varied, with a broad maximum at about 5 to 11 slices.

DBT enhanced tissue textures and spiculations. The spiculation and texture features from the margin are most important for characterizing malignant masses on DBT. More cases are being accrued for further analysis.

Feature analysis provides useful information for image interpretation and CAD development in this new breast imaging modality.

Chan, H, Wu, Y, Sahiner, B, Zhang, Y, Moore, R, Kopans, D, Helvie, M, Hadjiiski, L, Way, T, et al, , et al, , Analysis of Mass Characteristics on Digital Breast Tomosynthesis (DBT) Mammograms: Application to Computer-aided Diagnosis.  Radiological Society of North America 2007 Scientific Assembly and Annual Meeting, November 25 - November 30, 2007 ,Chicago IL. http://archive.rsna.org/2007/5005320.html