A study on computer aided diagnosis systems for lung cancer detection using machine learning techniques

Abstract

Lung cancer is considered as the notable cancer because it claims more than a million lives every year. The early detection of cancer can be helpful in curing the disease completely. The requirement of techniques to detect the occurrence of cancer nodule in the early stage is very much essential. Computer Aided Diagnosis (CAD) is becoming one of the most popular and effective method for diagnosing many diseases including cancer. The modalities used for capturing the images are X-Ray, Computer Tomography (CT) scans and Magnetic Resonance Imaging (MRI) and among these CT is the standard for detecting pulmonary nodules. The main aim of this research is to provide a Computer Aided Diagnosis System for detection of lung cancer nodules from the Chest Computer Tomography images. The system can automatically detect the lung cancer nodules with reduction in false positive rates and it also minimizes the time taken by the radiologist for interpretation. The scheme proposed in this thesis aims to design a good CAD system for detection of lung cancer in five phases Extraction of Lung Region from Chest Computer Tomography Images; Segmentation of Lung Region; Feature Extraction from the Segmented Region; Formation of Diagnostic rules from the extracted features; Classification of malignant and benign nodules. In the first stage of this CAD system pure basic image processing techniques are used to extract lung regions. Fuzzy Possibilistic C-Means (FPCM) is used to get good segmentation results in a short time. For automatic detection of cancer nodules, Support Vector Machine (SVM) and Extreme Learning Machine (ELM) are utilized. SVM is used because of its simplicity and accuracy. The usage of SVM helps in better classification of cancer nodules. The main advantage of the ELM is its accuracy and short training time. Real time lung images are used for the experimental observation of the proposed approaches. The performances of the proposed approaches are evaluated based on their accuracy, sensitivity, specificity and classificat