Lung cancer is a widespread global disease and is the most often occurring kind of cancer. More than 85% of cases are diagnosed with non-small cell lung cancer (NSCLC), which has a five-year survival rate of fewer than 20% and a wide range of treatment options. Timely detection and accurate diagnosis can enhance life expectancy and potentially preserve the patient’s life. Precise classification of lung cancer subclasses is crucial in guiding medical professionals in making informed judgments on treatment options for lung cancer. This study aims to enhance the classification accuracy of NSCLC using CT scans by proposing a novel deep-learning architecture that integrates convolutional neural networks (CNNs) and data efficient vision transformers (DeiT). This paper presents a novel two-branch parallel model that combines the Transformer and Convolutional Neural Networks (CNN) to accurately classify NSCLC in CT images. The branch module of DeiT leverages the global receptive field to capture global information from the CT scans while reducing the model’s dependency on extensive training data, while the CNN focuses on capturing local data. The fusion-based architecture enables the integration of parallel systems across different branches, leading to the formation of a pattern that can identify different classes of lung cancer. The proposed model’s performance result is evaluated using a dataset which is publicly available on Kaggle repository. The framework achieved an accuracy of 96.10%, sensitivity of 96.01%, specificity of 98.70%, and F1-score of 95.93% for the subclass categorization of lung cancer. The proposed approach was also evaluated against current state-of-the-art methods and shown significant improvement for classifying lung cancer subtypes.