Rationale and Objectives
Although spiculation level of breast mass boundary is a primary sign of malignancy for the mass detected on mammograms, developing an automated computer scheme to detect mass spiculation level and quantitatively evaluating the performance of the scheme is a difficult task. The objective of this study is to (1) develop and test a new scheme to improve mass segmentation and detect mass boundary spiculation level, and (2) assess the scheme performance using a relatively large image dataset.
Materials and Methods
This fully-automated scheme includes three image processing steps. The first step applies the maximum entropy principle in the selected region of interest (ROI) after correcting the background-trend to enhance the initial outlines of the masses. The second step uses an active contour model to refine the initial outlines. The third step detects and identifies spiculated lines connected to the mass boundary using a special line detector. A quantitative spiculation index is computed to assess the degree of spiculation levels. To develop and evaluate this automated scheme, we selected 211 ROIs depicting masses that were extracted from a publicly available image database. Among these ROIs, 106 depict “circumscribed” mass regions and 105 involve “spiculated” mass regions. The scheme performance was evaluated using the receiver operating characteristic (ROC) analysis method.
The computed area under ROC curve when applying the scheme to the dataset is 0.701 ± 0.027. By setting up a threshold at spiculation index = 5.0, the scheme achieves the overall classification accuracy of 66.4% with 54.3% sensitivity and 78.3% specificity, respectively.
We developed a new computer scheme with a number of unique characteristics to detect spiculated mass regions and applied a simple spiculation index to quantify mass spiculation levels. Although this quantitative index can be used to classify between the spiculated and circumscribed masses, the results also suggest that automated detection of mass spiculation levels remains a technical challenge.