Abstract
We have applied novel computational image analysis algorithms todetect malignant masses in mammograms. Our analysis focuses on spiculatedlesions, which are particularly challenging for computer-aided detectionmethods. The algorithm uses the principle of locally-normalised correlationcoefficients to identify patterns of motifs representing a spiculated feature. Acombination of correlation maps indicating the maximum correlation of the motifat each position relative to the mammogram, and of the pattern of angles forwhich this maximum is observed, are used to locate spiculated lesions in a verifiedtest dataset. The test set of images has been annotated by an expert reader,and allows objective evaluation of computer-aided detection procedures. In ablind test using an automated procedure our method identified 54% of the lesionlocations in the set of test images. This initial blind testing and comparisonwith expert annotated images, representing a ground truth, indicates feasibilityfor our approach. Optimisation of the procedure is expected to yield improvedperformance.
Original language | English |
---|---|
Title of host publication | Breast Imaging: Lecture Notes in Computer Science 8539 |
Editors | Hiroshi Fujita, T Hara, C Muramatsu |
Place of Publication | Switzerland |
Publisher | Springer Nature |
Pages | 550-557 |
Number of pages | 8 |
Publication status | Published - Jun 2014 |
Event | International Workshop on Breast Imaging - Gifu, Japan Duration: 1 Jan 1824 → … |
Conference
Conference | International Workshop on Breast Imaging |
---|---|
City | Gifu, Japan |
Period | 1/01/24 → … |
Keywords
- Mammography
- Detection
- Cancer
- Correlation
- Image Analysis