Model-based Framework and Algorithm for the Detection and Annotation of Spiculated Masses on Mammography

Life Sciences : Diagnostics

Available for licensing


  • Alan Bovik, Ph.D. , Electrical and Computer Engineering
  • Mia Markey, Ph.D. , Biomedical Engineering
  • Gautam Muralidhar , Biomedical Engineering
  • Mehul Sampat, B.E. , Biomedical Engineering

Background/unmet need

Breast cancer manifests as various findings on mammography, including microcalcifications, masses (spiculated and non-spiculated), and architectural distortions. Spiculated masses are characterized by a pattern of radiating lines known as spicules that emanate from a central mass. Spiculated masses have a much higher risk of malignancy than non-spiculated masses and calcifications and hence it is crucial to detect spiculated masses early. But the detection of these cancers is a repetitive and fatiguing task. Only three or four out of a thousand examined cases are malignant, and thus an abnormality may be overlooked. As a result, radiologists fail to detect 10% to 30% of cancers.

Computer-aided detection (CADe) systems have been developed to assist radiologists in detecting signs of early breast cancer. These systems act as a second reader, thus eliminating the need for a second radiologist. However, studies have shown that CADe systems to perform consistently better on microcalcifications than on masses, and in particular, the detection performance on spiculated masses is not optimal. This can be attributed to the absence of image processing devices designed to explicitly identify and annotate spicules, as a result of which there exists no reliable mechanism to build models of spicules that can be used for classification of suspect cancerous locations identified by CADe systems.

Invention Description

This invention is a new model-based framework for the detection of spiculated masses on mammograms and an evidence-based active contour algorithm to explicitly annotate these spicules on mammography.

The detection algorithm:
a) enhances spicules through Spiculation Filtering and detects the spatial locations where the spicules converge
b) detects the central mass region of the spiculated masses, and
c) reduces the false positives due to normal linear structures.

The foundation of this algorithm is strong, as all the parameters are based on actual physical properties of spiculated masses measured by experienced radiologists. The algorithm, when tested on a set of 100 challenging images from the publicly available DDSM database, showed a sensitivity of 88% at 2.7 FPI (sensitivity is the fraction of regions marked as suspicious that are actually lesions and FPI (false positives per image) is the number of regions marked per image that are not lesions). This technique aims to find the highest risk abnormalities and will be a useful aid to radiologists in detecting breast cancer.

Additionally, this invention comprises of a new image processing device we call Snakules that has been designed to explicitly annotate spicules on mammograms. Starting from a natural set of automatically detected candidate points, we deploy snakules that consist of converging open-ended active contours, also known as snakes. The set of convergent snakules (snakes that seek spicules) have the ability to grow and adapt to the true spicules in the image

Observer studies involving experienced radiologists to evaluate the performance of snakules demonstrate the strong potential of the algorithm as an image analysis technique to improve the specificity of CADe algorithms and as a CADe prompting tool.


  • Employs a model-based, evidence-based approach
  • Captures the exact shape of spicules on mammograms
  • The algorithm is modular and could be easily integrated with the existing CADe Systems, and Snakules is a fully software-based implementation that is easy to use in conjunction with existing CADe systems
  • New knowledge about the physical properties of spiculated masses or normal tissue structures can be easily incorporated into the framework of this algorithm
  • Design is easily extensible to images acquired using forthcoming 3D breast imaging modalities (e.g., digital breast tomosynthesis)


  • Enhancement of linear structures via filtering in the Radon Domain
  • A new class of filters, called spiculation filters, that were specifically designed to detect locations at which linear structures converge based on measurements of spiculated masses made by radiologists
  • Algorithm is evidence-based, using physical measurements of spiculated masses collected from mammograms to detect candidate points from which the snakes originate
  • Snakules grow by the process of curve evolution and motion that optimizes the local matching energy
  • Explicit models of normal structures used to reduce the number of false positive detections

Market potential/applications

Computer-Aided Detection (CADe) mammography systems

Development Stage

Lab/bench prototype

IP Status

Web Links