LUNG CANCER DETECTION USING DEEP LEARNING APPROACHES ON CT IMAGES

Abstract

This research aims to design and develop an automated system for lung cancer detection using deep learning methods applied to Computed Tomography (CT) images. Early and precise diagnosis of lung cancer is a must for better patient outcomes since lung cancer remains among the leading causes of death from cancer worldwide. The Lung Cancer Detection model described in this research utilizes a convolutional neural network architecture and is trained on the "The Iraq-Oncology Teaching Hospital/National Center for Cancer Diseases (IQ-OTH/NCCD) lung cancer augmented dataset," which comprises an extensive collection of computed tomography images with annotations. The deep learning model was trained on CT scan images categorized into Normal, Benign, and Malignant labels. These images underwent meticulous preprocessing and training using three deep learning algorithms: ResNet50, InceptionV3, and DenseNet121. The best performance, achieved by the custom CNN model, boasts an extraordinary classification rate of 98.30% and perfect precision, recall, and F1-score. Model classification for malignant or benign nodules will be based on imaging characteristics of the lung nodules. It employs a preprocessing approach similar to that used for CT images, data augmentation when necessary, and transfer learning to enhance the model’s performance, which is further evaluated based on model accuracy, sensitivity, and specificity. Limitations include the availability and overfitting of annotated datasets, which can be affected by variability in quality across different CT scans. In the future, more robust architectures should be developed, multi-model data should be integrated, and real-time detection features should be included to advance the research field of lung cancer diagnosis.