A Machine Learning Approach to Detect Lung Nodules Using Reinforcement Learning Based on Imbalanced Classification

Lung cancer is one of the fatal diseases affecting millions of people globally. Importantly, accuracy and rapid lung nodules detection on CT images are significant challenges. Doctors and radiologists' manual detection of lung nodules have low efficiency due to the variety of size, shape, and nodule’s location. Approaches based on machine learning play a fundamental and important role in the integration and analysis of these large and complex data sets. Considering the different viewpoints of learning methods and diverse lung cancer data, machine learning is an attractive and interesting method for researchers. This paper proposes a novel machine-learning approach to detect lung nodules using reinforcement learning based on imbalanced classification. Due to many healthy data, the ANN classifier tends towards this class rather than the suspicious one. To avoid this problem, we apply reinforcement learning to learn ANN parameters. The results obtained from our proposed model on the LIDC-IDRI data set showed better results than state of the arts on standard performance metrics such as IOU and HD. Also, according to accuracy metrics proposed method have the highest value of 94.91, compared to the previous best methods. Compared with advanced approaches, experimental results on the LIDC dataset indicated that the proposed model could detect various nodules and achieve the highest accuracy at the lowest time. Due to the nature of imbalanced data on most cancers, the proposed method can be adapted to other cancers.