Transformer-Based Hierarchical Model for Non-Small Cell Lung Cancer Detection and Classification

Abstract

Lung cancer is the leading cause of cancer-related deaths worldwide. Early diagnosis significantly improves the 5-year survival rate from 6% in patients with metastatic cancer to 60% in those with localized cancer. Histopathological examination is the gold standard for lung cancer diagnosis, but analyzing whole slide images (WSI) is time-consuming and prone to error for pathologists. This study aims to enhance the classification accuracy of non-small cell lung cancer (NSCLC) histopathological images by proposing a novel deep-learning architecture that integrates convolutional neural networks (CNNs) and vision transformers (ViTs). The model classifies NSCLC into three categories: normal, adenocarcinoma, and squamous cell carcinoma. CNNs are employed to capture local features, while ViTs are used to understand long-range relationships between image patches. We trained and validated our model on the LC25000 dataset, a benchmark dataset for NSCLC histopathology image classification. Our proposed model demonstrated superior performance, achieving an accuracy of 0.988, an F-1 score of 0.980, specificity of 0.991, recall of 0.982, and precision of 0.980, outperforming existing state-of-the-art methods. Additionally, our model achieved a low inference time of 1.816 ms, highlighting its potential for real-world applications where both accuracy and speed are critical. Our code is now publicly available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/ImranNust/LungCancerDetection</uri> to facilitate further research and validation of our findings.