A framework for lung oblique fissure segmentation and lung nodule detection in lung ct images

Abstract

The advances in computerized medical image analysis provide helpful tools for the diagnosis of various diseases and treatment process. Also, it provides the hidden characteristics of the medical images which are sometimes difficult to see with a naked eye. This is the reason that the application of medical image analysis have emerged in recent years. Nowadays, Computer Aided Diagnosis (CAD) system which is a computerized medical image analysis system plays a vital role in diagnosis as a second reader. This research work develops a CAD system for lung cancer diagnosis using Computer Tomography (CT) images of lungs scans with two important modules: (i) detection of oblique fissures and (ii) detection of lung nodules using various image processing and machine learning algorithms. The CAD system for the detection of oblique fissures in CT lung image consists of three steps; preprocessing, lung region extraction and oblique fissure detection. In the preprocessing stage, the lung structures are enhanced using morphological algorithm and the noise in lung images is filtered using Wiener filter and. In next step, lung region are segmented using thresholding followed by background subtraction. Fissure region are identified using the Chan-Vese active contour model. Finally, the oblique fissure is segmented and then detected using regional rule-based approach. To detect the lung nodules, the input CT images are preprocessed to remove the noise initially. After noise removal, two-successive segmentation process is employed to obtain the lungs region only. Then, nodules and vessels are segmented using thresholding followed by morphological operations. The false positives in the lung regions are reducedby using rule based filtering. Finally, using Faster Region-based Convolutional Neural network (FR-CNN) with batch normalization inception model, the nodules are detected. The evaluation of the CAD system is carried out using CT lung images from Early Lung Cancer Action Program (ELCAP) database and Lung Image Database Consortium (LIDC) database. The oblique fissure detection algorithm is tested on 130 lung CT scans from LIDC dataset. The obtained RMS error for the detection oblique fissure in the left lung is 2.968 pixels whereas it is 3.221 for right lung. For the 30 CT scans from ELCAP dataset, the obtained RMS errors are 1.357 and 3.043 pixels for left and right lung respectively. As there is no clear annotation available in the ELCAP database, the performance of lung nodule detection algorithm is tested on LIDC database images. From the results of CAD system, it is observed that this system provides an accurate and early detection of lung nodule with an average precision of 96.75% on LIDC database and also segments the oblique fissures accuratel newline