Improved Lung Nodules Detection with Active Contour Method and Machine Learning Based Classification

Computed tomography (CT) scan is one of the widely used medical diagnostic procedures to generate section images of the lung. CT scan employs computer-processed X-rays that rotates around the patient’s body, and stacks the tomographic images in three dimensional (3D) form for an easier identification of abnormalities. Due to the high mortality rate from lung cancer, CT scan alone may be insufficient for early detection of lung nodules. In order to improve detection accuracy, we propose a Computer Aided Detection (CAD) system to identify lung nodules. The system uses active contour method to build three-dimensional model from the scans. Then we can get the 3D lung model without trachea, bronchus, and nodules. By using the reversed out 3D model and threshold method, we can have the 3D models of trachea, bronchus, and nodules. To identify the nodules from the others, we use machine learning method to classify the features from the 3D modules. The proposed system shows 72.22% sensitivity for nodule within 8mm to 90mm with 0.16 false positive per dataset