|
|
 |
| |
|
|
|
Effective Extraction of Gabor Features for False Positive Reduction and Mass Classification in Mammography |
|
|
|
PP: 397-412 |
|
|
|
Author(s) |
|
|
|
Muhammad Hussain,
Salabat Khan,
Ghulam Muhammad,
Iftikhar Ahmad,
George Bebis,
|
|
|
|
Abstract |
|
|
| Digital mammography is considered to be the most effective imaging modality for early detection of breast cancer. Masses
and microcalcifications are two early signs of breast cancer. For the detection of masses, segmentation of mammograms results in ROIs
(regions of interest) which not only include masses but suspicious normal tissues as well, which lead to false positives. The problem
is to reduce the false positives by classifying ROIs as masses and normal tissues. In addition, the detected masses are required to be
classified as malignant and benign. We address these two problems using textural properties of masses. Gabor filter bank is used in a
novel way to extract the most representative and discriminative textural properties of masses present at different orientations and scales.
SVM with Gaussian kernel is employed for classification. The method is evaluated over 1024 (512 masses and 512 normal) ROIs
extracted from DDSM database. Experiments have been performed with different parameter settings to find the best set of parameters.
Gabor filter Banks with different choices of orientations (3, 5, 6, 8) and scales (2, 3, 4, 5) have been tested on 4 ROI resolutions
(64×64, 128×128, 256×256, 512×512). For the first problem i.e. to classify ROIs as masses and normal tissues, the best result (Az =
0.96±0.02) is obtained when Gabor filter bank with 5 orientations and 3 scales and RIOs with size 512×512 is used. Gabor filter bank
with 8 orientations and 5 scales on mass ROIs of size 128×128 gives the best result (Az = 0.87±0.05) for the second problem (i.e.
to classify mass ROIs as benign and malignant). Comparison with state-of-the-art methods reveals that the proposed method performs
better than the existing methods. |
|
|
|
|
|
|
|
