A lung tumor detection technique using gradient vector flow algorithm

Abstract

Lungs are a pair of spongy, air-filled organs that are a part of the Human Respiratory System. Cancer is a disease where the cell growth in a body goes out of control. Such uncontrollable abnormal cell growth in the lungs leads to lung cancer. Lung tumors can be categorized as small cell lung cancer and non – small cell lung cancer depending on the size of the cell as perceived under a microscope. More people are affected by non-small lung cancer than small-lung cancer. Tumors, which are detected at a prior stage have more likelihood of getting treated at a faster period, however, if left undetected or undiagnosed for a long period, it could lead to various obstructions. As a fact, the traditional methods used for diagnosing are time-consuming and have more probability of chances of errors. Thus a non – invasive method of diagnosing has been studied and discussed. In the proposed study, PET-CT images have been collected from various online databases and processed using Gradient Vector Flow (GVF) Algorithm. As a result of the study, the proposed gradient vector flow algorithm provides a precise segmentation in the lung images. Statistical features like mean, kurtosis, skewness, standard deviation, have been studied and compared between the normal and abnormal images. Thus a most precise and fast programmed method has been implemented to segment the lung tumor images using a gradient vector flow algorithm. Machine learning classifiers like KNN, Naïve Bayes, multilayer perceptron and random forest were analyzed. It was observed that the accuracy value of KNN classifier is 91.8%, Naive Bayes Classifier is 91.9%, Multilayer Perceptron is 89.1% and Random Forest Classifier is 87.9%.