Classification of non-small cell lung cancers using deep convolutional neural networks

Lung cancer is a major cause of cancer-related deaths worldwide, and early detection is crucial in reducing mortality rates. To aid in this effort, researchers have been exploring various deep-learning techniques to enhance computer-aided systems that utilize computed tomography in lung cancer screening. One such technique is transfer learning, which allows for the use of pre-trained models to reduce the need for extensive training data. However, deep convolutional neural networks (DCNNs), which are commonly used in deep learning, can be challenging to train due to over-fitting, and effective training requires substantial amounts of data. To address these limitations, the authors propose a dual-state transfer learning method using a deep CNN-based approach. This method aims to develop an efficient training model that reduces variance and avoids over-fitting, while accurately classifying and detecting lung cancer types such as adenocarcinoma, squamous cell carcinoma, and small cell carcinoma, using CT-scanned chest images. In order to achieve effective results, the authors utilized pre-trained models such as the DCNN, VGG16, Inceptionv3, and RestNet50. Metrics like the f1-score, recall, precision, and accuracy were used to evaluate the performance of the proposed model. During training, the ResNet50 model achieved an accuracy of 94% using dual-state transfer learning, while during validation and testing it achieved 92.57% and 96.12% accuracy, respectively. In classification tasks, the DSTL model based on Deep CNNs also surpassed state-of-the-art models. To summarize, the precision and effectiveness of screening and detecting lung cancer can be improved by utilizing dual-state transfer learning methods and deep CNN-based approaches.