Attention-based deep learning for analysis of pathology images and gene expression data in lung squamous premalignant lesions

The preprint operates within the dominant paradigm of using deep learning, specifically attention-based models, for image analysis and multi-modal data integration in computational pathology. It adheres to the established practices of training and validating models on large datasets, using metrics like AUROC and AUPR for evaluation, and exploring the biological relevance of the model's findings.

This preprint falls squarely within the realm of normal science. It applies existing deep learning techniques to a specific problem within the field of lung cancer research – improving the diagnosis and understanding of premalignant lesions. While the authors introduce a novel framework for combining whole-slide images and gene expression data, they do so within the accepted methods and norms of the field. There's no challenge to existing paradigms or proposals of radical new theories. The work aims to improve and refine current diagnostic tools using established deep learning methodologies, contributing incrementally to the body of knowledge within the existing paradigm. Therefore, the classification is clearly 'Normal Science'.