An adaptive teaching learning based optimization technique for feature selection to classify mammogram medical images in breast cancer detection

Feature Selection is the vibrant phase in Computer Aided Diagnosis Systems (CAD) to classify breast cancer medical images. Mammogram images are best for breast cancer screening which helps in early detection of the disease in the women. The main objective is to select the better features to increase the classification performance. To achieve this, we used a 4-step process: 1. preprocessing 2. feature extraction, 3. feature selection, 4. classification. Initially, the medical images are acquired and preprocessed using Contrast-Limited Histogram Equalization (CLAHE) and the features are retrieved using Advanced Gray-Level Co-occurrence Matrix (AGLCM) which extracts texture, shape and intensity-based features. Then feature selection is applied to obtain the better features. Toachieve this, we proposed a new feature selection technique called Weighted Adaptive Binary Teaching Learning Based Optimization (WA-BTLBO) and fitness function used is the classification accuracy. The selected features are trained and tested using XGBoost classifier and the results are compared with other classifiers namely K-Nearest Neighbor (KNN), Random Forest (RF), Artificial Neural Networks (ANN), andSupport Vector Machines (SVM). The experiments are done using publicly available mammogram medical images namely Mammographic Image Analysis Society (MIAS). Theresults show that WA-BTLBO with XGBoost classifier is superior to other feature selection techniques namelyParticle Swarm Optimization (PSO) and Binary TLBO (BTLBO) and other state-of-art methods in classifying MIAS mammogram images into normal or abnormal. This paper helps the physiologists and radiologists to detect the breast cancer in women so that life span of the patient can be increased. In future, we extend our work to use other metaheuristic methods like firefly algorithm, biogeography-based optimization and other algorithms to select the optimal features and also to apply on large databases and on other types of diseases.