Feature Selection Using Flower Pollination Optimization to Diagnose Lung Cancer from CT Images

A segmentation approach using snake splines and a flower pollination based feature selection are suggested in this work to diagnose lung cancer from Computed Tomography scan images. The proposed system segments the lung tissues including the diseased portion using a prior shape model generated using snake splines. The model generation includes the usage of control points to construct a snake spline model. The model is scaled on to the diseased lung tissues so as to segment the lung region along with its boundary. The level of scaling varies depending on the size of reference and diseased lung image and hence requires certain affine transformations to fit. From the segmented lung, the cancerous regions which are the region of interests (ROIs) are extracted. Features are extracted from these ROIs. A binary flower pollination algorithm is used to select the relevant features in wrapper approach for further classification using SVM classifier. In this work, 514 ROIs are extracted among which 414 ROIs are used for training and 100 for testing in relation to 80:20 Pareto principle. 33 features are selected by binary flower pollination algorithm from 56 extracted features. The proposed approach achieved an accuracy of 84% using SVM classifier. The results obtained when compared with similar algorithms, either selected more than 33 features or yielded a lower accuracy.