Lung Nodule Detector and Classifier Tool

Abstract

Lung cancer is the number one cause of death in the Philippines and in the world. Multi-detector CT scanners provide opportunity to examine thin-section CT images, which improves reader detection of focal findings and characterization of nodules. However, the sensitivity of manual detection of cancerous and non-cancerous lung nodules is reported to be 70-75% only. Lung Nodule Detector and Classifier Tool is decision support software that aims to aid the health professionals in detecting and classifying lung nodules. LNDCT uses algorithms which include diffusion, binarization, wavelet edge detection, and morphological operations to automatically segment the nodules. The tool uses LibSVM, an open source support vector machine software in classifying malignant from benign nodules. Three testing sets were used to test the accuracy of detection and classification using LNDCT. Set A contains lung images where most of the sizes of the cancerous nodules are greater than 200 pixels. Set B contains a combination of different sizes of cancerous nodules. Set C contains 400 training and 200 testing random data which came from pre-extracted features computed using LNDCT. The accuracy of the system is reported to be 93.87%, 81.33%, and 88.57% for sets A, B and C respectively. LNDCT is a tool which can help health professionals in detection, classification, and feature computations of lung nodules.