Classification of abnormalities in digitized mammograms using Extreme Learning Machine

Digital mammography is a preferred method for early detection of breast cancer. However, in most cases, it is very difficult to distinguish benign and malignant masses without a biopsy, hence, misdiagnosis is always possible. In this paper, the Extreme Learning Machine (ELM) algorithm is used to classify the suspicious masses in digitized mammograms available in the Mini-MIAS database. As selection of features is critical in efficient classification of mammograms, a study was conducted to identify the best features that make a clear distinction between the malignant and benign masses. Then, the recently developed Extreme Learning Machine (ELM) based classifier is used to classify the benign and malignant cases. It is observed that the performance of the batch learning ELM algorithm is dependent on the number of hidden neurons, and the magnitude of input weight initialization. An extensive study is conducted to identify the best number of neurons to achieve the best training and testing classification performance. The performance results are then compared with the classification results, available in the literature. The performance study results show that the classification efficiency of the ELM classifier is better than the other existing algorithms, for the mammogram classification problem of the database considered for this study.