Research Article

The Impact of Pixel Resolution, Integration Scale, Preprocessing, and Feature Normalization on Texture Analysis for Mass Classification in Mammograms

Table 1

Summary of texture analysis methods that have been used to analyze mammograms.

Method Extracted features Utilized classifiers Purpose

[6] Local binary pattern (LBP) Support vector machines (SVMs) Classification of ROIs into mass/normal
[7] Histogram of oriented gradients (HOG) SVM Classification of ROIs into mass/normal
[11] Haralick’s features (HAR)-nearest neighbour (-NN) Microcalcification classification
[8] Gabor filters (GF) Threshold-based approach Breast cancer detection
[29] Grey levels, texture, and features related to independent component analysis Neural network (NN) Classifying ROIs into normal/abnormal
Classifying ROIs into benign/malignant
[30] A set of texture features SVM Mass detection
[31] Ripley’s function texture measures SVM Detection of breast masses
[9] Texture features derived from concurrence matrix NN Microcalcification classification
[32] A set of texture features -NN, SVM, random forests, logistic model trees, and Naive Bayes Lesion classification
[10] HAR Bayesian classifier
Fisher linear discriminant
Study the effect of pixel resolution on the performance of texture methods
[3] LBP, robust LBP, centre symmetric LBP, fuzzy LBP, local grey level appearance, LDN, HOG, HAR, and GF -NN, linear SVM, nonlinear SVM random forest, and Fisher linear discriminant analysis (FLDA) Finding the best combination among the texture methods to classify ROIs into mass/normal
[33] Local ternary pattern and local phase quantization SVM Classifying tumors into benign/malignant
[34] Novel sets of texture descriptors extracted from the cooccurrence matrix SVM Six medical datasets were used for validation, one of them for breast cancer
[35] Texture analysis techniques based on the cooccurrence matrix and region-based approaches SVM 15 datasets were used for validation, one of them for breast cancer
[36] HOG, dense scale invariant feature transform, and local configuration pattern SVM, -NN, FLDA, and decision tree Classifying ROIs into normal/abnormal
Classifying ROIs into benign/malignant
[37] Curvelet moments -NN Classifying ROIs into normal/abnormal
Classifying ROIs into benign/malignant