A novel approach for tumor segmentation for lung cancer using multi-objective genetic algorithm and connected component analysis

Abstract

Treatment in oncology relies heavily on imaging. Automatic identification of cancerous tumors is a challenging and time-taking task. This paper presents a multi-objective genetic algorithm for calculating optimal threshold value for binarization of CT lung images and segmentation of the tumor region using connected component analysis. The threshold determination method is tested on two public datasets: Data Science Bowl: Stage 1 and NSCLC lung cancer dataset. The results obtained are compared with other genetic algorithm approach and traditional Otsu’s method. The proposed method gives better threshold value compared to Otsu’s method and outperforms other genetic algorithm approach both in terms of threshold value and computation time. Connected component analysis is done on the labeled connected component in the binarized image. Based on predetermined geometric measurements, the components are marked as tumors and non-tumors. This method when applied to 240 cancerous images, correctly identified tumors for 205 images while showed no result where there are no tumors in the image.