Deep Learning-Based Segmentation Of Lung Images For Accurate Diagnosis

Abstract

As lung cancer continues to be the foremost cause of death worldwide, it is crucial to explore various diagnostic approaches such as computed tomography, magnetic resonance imaging, and radiography. We aim to develop an advanced image processing method that utilizes segmentation algorithms to distinguish between CT scan images of lung cancer. To assess the technique’s effectiveness, we will compare three segmentation methods to the "ground truth" established by an oncologist, utilizing accuracy, precision, recall, and F-score tests. Our methodology involves information gathering, image segmentation, and area enlargement as the primary image processing techniques. Ultimately, the goal is to provide a dependable and efficient diagnostic tool for lung cancer through research. Various image segmentation techniques, such as k-means clustering, Otsu's thresholding, and watershed segmentation, were successfully utilized to separate lung images. The region growth method was then applied to measure the lung area accurately. A performance study was conducted to evaluate the efficiency of the segmentation algorithm. Processing medical MRI or CT scans present unique challenges in computer vision, as it requires careful consideration of spatial information to ensure accurate alignment and orientation of volumes. To streamline medical image computing pipelines and enable more deep learning research, the TorchIO platform was developed.