Background: Breast cancers are conventionally the leading cause of
cancer-related deaths amongst women. For the reduction of death rates by earlier
identification of carcinogenic areas, mammogram images are utilized. Computer aided
diagnosis plays an important role in screening of the mammograms. Mammography
is an efficient as well as feasible method for the detection of breast cancers, especially
minute tumours. Efficient performance of these tools is dependent on the efficacy of
the classifier algorithms. In this work, feature selection techniques are proposed to
improve the efficacy of the classifiers.
Method: The major phases in diagnosing breast cancers are features extraction and
selection. Detecting tumours naturally requires extraction of features as well as their
classification. This work uses Pseudo Zernike Moments and Gaussian Markov Random Field (GMRF)
for feature extraction, Binary Shuffled frog algorithm, Information Gain (IG) and Binary Particle Swarm
Optimization (PSO) for feature selection and C4.5, Random tree, Adaboost as classifiers.
Results: The use of feature selection techniques for successfully selecting relevant feature subset
improves the classification performance. The proposed wrapper based Binary Shuffled Frog Algorithm
improves the detection of breast cancer mass in mammograms.
Conclusion: The work focuses on improving classification performance through feature selection. The
experimental results demonstrate the efficacy of the proposed feature selection method. It is observed that
the feature selectors are necessary to improve the efficiency of the classifiers. It is observed that among
the various classification techniques, C4.5 outperforms other algorithms achieving the highest accuracy.