Enhancing Lung Segmentation Through Preprocessing of Medical Data Using Convolutional Neural Networks

Lung cancer is the leading cause of cancer-related death for women and the second leading cause of cancer-related death for men, according to a report by the World Health Organization in 2020. Detecting lung cancer through a three-dimensional X-ray of a patient's lungs, or a chest X-ray, is a common practice, but it can lead to errors caused by the human factor. To address this issue, advanced algorithms, and CAD technologies can be used to detect lung cancer. A network model can be trained using deep structural learning to detect details and deviations in the X-ray image. However, creating an accurate model using this technique is both time-consuming and advanced, and the model must be trained correctly. Despite the extensive research on various network architectures, the preprocessing technique of lung segmentation and its influence on such models remains largely unexplored. Thus, in this study, we aim to investigate the effect of applying lung segmentation to the training and testing data used by the CNN model on accuracy and loss. To achieve this, we developed two CNN models, one with and one without lung segmentation, respectively. The models were evaluated and compared, and our results demonstrate that lung segmentation can mitigate overtraining, resulting in a significant improvement in accuracy. We anticipate that this study will pave the way for further research in similar problem areas.