A modified watershed segmentation algorithm for automated lung cancer lesion detection in CT scan images / Nur Najihah Sofia Mohd Marzuki

Lung cancer is one of the leading cancers among both men and women and ranks among the most dangerous and life-threatening diseases worldwide. Computed Tomography (CT) imaging is a common method for detecting lung cancer. In the era of advanced computer technology, Computed Aided Diagnosis (CAD) has gained prominence particularly in medical applications such as diagnosing lung cancer. In medical applications, such as diagnosing lung cancer, CAD systems are adopted, utilizing lung CT images as input and various algorithms to assist doctors in image analysis and decision-making. A key problem addressed in this study is the challenge of accurately distinguishing between lesion and non-lesion areas in lung CT images. This difficulty underscores the necessity for an effective segmentation method, a critical aspect highlighted throughout this research. Therefore, the main aim of this research is to establish a segmentation method suitable for the automated detection of lung cancer in CT scan images. To achieve this goal, the research is divided into three parts: image pre-processing, segmentation to detect lung cancer lesions, and feature extraction for lung cancer identification. The dataset utilized comprises images from The Cancer Imaging Archive (TCIA Images) as a benchmark and the Imaging Department at the Advanced Medical and Dental Institute (AMDI), USM, Bertam, Pulau Pinang, Malaysia, involving 50 subjects. The image processing technique is implemented using MATLAB software, with the input CT scan images in grayscale. Initially, the images undergo thresholding, image filtering, and enhancement processes to obtain clearer lung area images. This pre-processing stage is essential to improve image quality, removing unwanted information that may obscure important features and reducing distortion or noise. The next crucial stage is the segmentation, where modified watershed is used to demarcate the lung region in the CT scan images. The performance of the segmentation process in conventional and modified watershed segmentation technique for detecting lung lesion, which produce average F-Score is 97.80% and 99.09%, respectively. The outcome of this research is highly valuable for doctors in determining appropriate treatments for patients and diagnosing lung cancer and nodules from the images. Implementing CAD on CT scan images can potentially aid doctors in diagnosing lung cancer and nodules effectively.