Effective lung cancer detection in CT images using segmentation and Convolutional Neural Networks

Lung cancer still retains its position among the deadliest diseases in the world, as the WHO considers lung cancer a cause of about 7.6 million deaths every year. Such a figure is expected to increase in the future, with the growth rate indicating that the figure may reach seventeen million by 2030. It is important to diagnose lung cancer in its early stage because this has implication to the success of treatment. Diagnostic methods such as Magnetic Resonance Imaging, Isotope Imaging, X-rays and Computer Tomography are quite popular; interpreting CT images at an early stage of disease detection is not easy. Recent work that employs convolutional neural networks with image segmentation toward the detection of features on endoscopy images is a possibility for enhanced sensitivity. Here in this study, we use CNN models along with image segmentation techniques to diagnose the CT scans, which helps the clinicians to detect lung cancer easily by visualizing the selective area and determining the important features within the scan. Being a deep learning model, CNN is particularly successful in identifying subtleties and complex textures in the medical data. This research aims to apply an end-to-end CNN framework to examine feature relevancy extraction against state-of-the-art conventional, as well as other traditional CAD systems. Our proposed model exhibits a 93% accuracy, and hence, better results in lung cancer diagnosis. The proposed work concludes that CNN-based segmentation methods may further improve early detection, potentially providing a major boost to diagnostic processes. Future work might consist of extending the presented approach to other imaging modalities and adapting the model architecture to run in real-time in clinical settings.