Open Access Open Access  Restricted Access Subscription Access

A Modified Binary PSO Based Feature Selection for Automatic Lesion Detection in Mammograms


Affiliations
1 Department of Computer Applications, BPC College, Piravom, India
2 Department of Computer Science, Government College, Nedumangad, India
3 Department of Imageology, Regional Cancer Center, Thiruvananthapuram, India
 

This paper presents an effective feature selection method that can be applied to build a computer aided diagnosis system for breast cancer in order to discriminate between healthy, benign and malignant parenchyma. Determining the optimal feature set from a large set of original features is an important preprocessing step which removes irrelevant and redundant features and thus improves computational efficiency, classification accuracy and also simplifies the classifier structure. A modified binary particle swarm optimized feature selection method (MBPSO)has been proposed where k-Nearest Neighbour algorithm with leave-one-out cross validation serves as the fitness function. Digital mammograms obtained from Regional Cancer Centre, Thiruvananthapuram and the mammograms from web accessible mini-MIAS database has been used as the dataset for this experiment. Region of interests from the mammograms are automatically detected and segmented. A total of 117 shape, texture and histogram features are extracted from the ROIs. Significant features are selected using the proposed feature selection method.Classification is performed using feed forward artificial neural networks with back propagation learning. Receiver operating characteristics (ROC) and confusion matrix are used to evaluate the performance. Experimental results show that the modified binary PSO feature selection method not only obtains better classification accuracy but also simplifies the classification process as compared to full set of features. The performance of the modified BPSO is found to be at par with other widely used feature selection techniques.

Keywords

Binary Particle Swarm Optimization, Feed Forward Artificial Neural Networks, Feature Selection, K-Nearest Neighbour.
User
Notifications
Font Size


  • A Modified Binary PSO Based Feature Selection for Automatic Lesion Detection in Mammograms

Abstract Views: 488  |  PDF Views: 212

Authors

K. U. Sheba
Department of Computer Applications, BPC College, Piravom, India
S. Gladston Raj
Department of Computer Science, Government College, Nedumangad, India
D. Ramachandran
Department of Imageology, Regional Cancer Center, Thiruvananthapuram, India

Abstract


This paper presents an effective feature selection method that can be applied to build a computer aided diagnosis system for breast cancer in order to discriminate between healthy, benign and malignant parenchyma. Determining the optimal feature set from a large set of original features is an important preprocessing step which removes irrelevant and redundant features and thus improves computational efficiency, classification accuracy and also simplifies the classifier structure. A modified binary particle swarm optimized feature selection method (MBPSO)has been proposed where k-Nearest Neighbour algorithm with leave-one-out cross validation serves as the fitness function. Digital mammograms obtained from Regional Cancer Centre, Thiruvananthapuram and the mammograms from web accessible mini-MIAS database has been used as the dataset for this experiment. Region of interests from the mammograms are automatically detected and segmented. A total of 117 shape, texture and histogram features are extracted from the ROIs. Significant features are selected using the proposed feature selection method.Classification is performed using feed forward artificial neural networks with back propagation learning. Receiver operating characteristics (ROC) and confusion matrix are used to evaluate the performance. Experimental results show that the modified binary PSO feature selection method not only obtains better classification accuracy but also simplifies the classification process as compared to full set of features. The performance of the modified BPSO is found to be at par with other widely used feature selection techniques.

Keywords


Binary Particle Swarm Optimization, Feed Forward Artificial Neural Networks, Feature Selection, K-Nearest Neighbour.

References