Image feature extraction for mass detection in digital mammography: Influence of wavelet analysis

Abstract

Rationale and objectives. The objective of this work is to evaluate the importance of image preprocessing, using multiresolution and multiorientation wavelet transforms (WTs), on the performance of a previously reported computer assisted diagnostic (CAD) method for breast cancer screening, using digital mammography. Method: An analysis of the influence of WTs on image feature extraction for mass detection is achieved by comparing the discriminant ability of features extracted with and without the wavelet‐based image preprocessing using computed ROC. Three indexes are proposed to assess the segmentation of the mass area with comparison to the ground truth. Data was analyzed on the region‐of‐interest (ROI) database that included mass and normal regions from digitized mammograms with the ground truth. Results: The metrics for the measurement of segmentation of the mass clearly demonstrated the importance of image preprocessing methods. Similarly, the relative improvement in performance was observed in feature extraction based on the evaluation of the ROC curves, where the Az values are increased, for example, from 0.71 to 0.75 for a pixel intensity feature and from 0.72 to 0.85 for a morphological feature of the Normalized Deviation of Radial Length. The improvement, therefore, depends on the feature characteristics, being large for boundary‐related features while small for intensity‐related features. Conclusion: The use of image preprocessing modules using wavelet transforms results in a significant improvement in feature extraction for the previously proposed CAD detection method. We are therefore exploring additional improvement in wavelet‐based image preprocessing methods, including adaptive methods, to achieve a further improvement in performance and an evaluation on larger image databases.

Evaluating the effect of a wavelet enhancement method in characterization of simulated lesions embedded in dense breast parenchyma

“…Application of the method on simulated lesion sample demonstrated statistically significant improvement of lesion contour sharpness/unsharpness characterization, as proven by area under ROC curve and CIs. Wavelet enhancement improvement is in agreement with our previous pilot preference study [23], reported lesion detection and classification tasks [13,20], as well as with automatic mass detection [37].…”

Prediction of near-term risk of developing breast cancer using computerized features from bilateral mammograms

“…For unilateral feature analysis, the proposed scheme computed the following groups of features including (1) morphological features: circularity, normalized deviation of radial length, the area of the extracted region in pixels, boundary irregularity factor; (2) textural features: the number of spiculations, the average spiculation length. The description of these features can be found in our previous papers [11][12][13].…”

“…Features (7)(8)(9) are the ratios between the average value of pixels under thresholds of 25%, 50%, and 75% of the maximum pixel value and the average pixel value of the whole breast area. Features (10)(11)(12) are the ratios between the number of pixels under each threshold and the total number of pixels of the entire segmented breast area. Feature 13 is the ratio between subtraction of the average value of pixels under 75% threshold minus average value of whole breast area and subtraction of average value of whole breast area minus average value of pixels under 25% threshold.…”

“…However, due to the large variability in the depiction of breast abnormalities, the overlapping dense fibroglandular tissue on the projection images and the low cancer prevalence in the screening environment, both detection sensitivity and specificity of screening mammography are relatively low [6][7][8]. To help radiologists improve detection and diagnosis performances in reading and interpreting screening mammograms, a great amount of research has been conducted to develop CAD systems or schemes including our work on developing a variety of two-dimensional computerized image analysis algorithms optimized to enhance the performance of the traditional CAD systems during the last two decades (e.g., [9][10][11][12][13]). Currently, a number of commercialized CAD schemes, including the one originally developed in our group and then being licensed to Carestream Health, Inc [14,15], are widely used in the clinical practice to assist radiologists in reading and interpreting mammograms to date.…”

Segmentation of ultrasound images by using an incremental self-organized map

“…In order to construct realistic classifiers, the features that are sufficiently representative of the physical process must be searched. In the literature, it is observed that wavelet transform [1], co-occurence matrix [2], and Gabor transform [3] are used to extract features from images. These feature extraction methods increase the computational time of the classification process [4].…”