Improved cervix lesion classification using multi-objective binary firefly algorithm-based feature selection

Abstract

Cervical cancer is one of the vital and most frequent cancers, but can be cured if correctly diagnosed. This work is a novel effort towards developing a methodology for effective characterisation of cervix lesions that may assist radiologists in the diagnostic process by providing a reliable and objective discrimination of benign and malignant lesions in contrast enhanced CT-Scan images. Feature selection, which is a key stage in building such efficient classification models, is NP-hard; where, randomised algorithms do better. Since, firefly algorithm is an efficient biologically inspired randomised algorithm; here it has been utilised for optimal feature selection. This paper presents a multi-objective binary firefly algorithm for wrapper-based feature selection and utilises the selected feature subset for improved classification of cervix lesions. For experiments, contrast enhanced CT-Scan images of 22 patients have been used, where all lesions had been recommended for surgical biopsy by specialists. For characterisation of lesions, grey-level cooccurrence matrix-based texture features are extracted from two-level decomposition of wavelet coefficients. The objective function is designed to minimise the classification error and feature subset length both; making it multi-objective. With 94% accuracy in lesion classification, it has superior performance and greatly reduced execution time than multi-objective genetic algorithm-based feature selection.