Imbalanced class problem analysis for lung cancer detection using convolutional neural networks

The most common cancer to be diagnosed is lung cancer, especially in men. The early detection of lung cancer allows for effective treatment, perhaps saving lives. The convolution neural network (CNN)-based deep learning methodologies are commonly used to detect lung cancer from CT images. These methodologies result in an imbalanced class distribution. To resolve the problem of an imbalanced class distribution, we propose three models: synthetic minority oversampling technique (SMOTE) analysis, a class-weighted approach, and data augmentation on CNN models. In this chapter different CNN based approaches, ResNet50, AlexNet, and DenseNet-121, are used to classify lung CT images to detect cancer. In this work, the CT image dataset from the Iraq-Oncology Teaching Hospital/National Centre for Cancer Diseases (IQ-OTH/NCCD) is used to evaluate the proposed method. The pre-processing of data is performed and applied to different CNN models. DenseNet-121 achieved an accuracy of 99.4% with SMOTE analysis. The classweighted and data augmentation approaches achieved accuracies of 97.4% and 95%, respectively, whereas ResNet50 and AlexNet obtained accuracies of 86.05% and 96.5%, respectively. The experimental results show that DenseNet-121 with SMOTE analysis achieved the highest accuracy. The obtained results show that the proposed method outperforms the state-of-the-art approaches.