Comparative Analysis of Learning Algorithms for Lung Cancer Identification

Abstract

Lung Cancer detection making use of medical imaging is still a challenging task for radiologist. The objective of this research is to classify the types of lung tumours for extracted and selected features using learning algorithms. In this paper, an experimental study is conducted on 100 cases of lung cancer to evaluate the performance of learning classifiers (DNN, SVM, Random Forest, Decision Tree, Naïve Bayes) with different medical Imaging (DICOM) features to identify the two types of Lung cancer (Benign and Malignant). The proposed methodology intends to automate the entire procedure of diagnosis by automatically detecting the tumor, measuring the required values such as diameter, perimeter, area, centroid, roundness, indentations and calcification. Experiment is conducted in to two phases: In the first phase, identify the most significant feature used in lung cancer analysis by CT scan and perform the mapping to computer related format. In the second phase, feature selection and extraction is performed to machine learning algorithms. To evaluate the performance of classifiers in term of classification accuracy and improving the false positive rate, every stage of evolution is divided into four different phases: single phase module, single slice testing, series testing and testing of learning algorithms. Experimental results show significant improvement in false positive rate up to 30% for both Benign and Malignant. Whereas, Deep Neural Network (DNN) demonstrate high values in term of classification accuracy in comparison with other classifiers. The proposed methodology for lung cancer detection system having a potential to reduce the time and cost of diagnosis procedure and use for early detection of lung cancer. Keywords: Cancer Identification, Deep Neural Network, Learning Algorithm, Medical Imaging