Toward a unified computational representation of lung cancer: from radiomics and pathomics to multimodal AI.

Authors

Abstract

Artificial intelligence (AI) is changing how lung cancer is characterized by enabling quantitative analysis of both radiologic and pathologic images. Radiomics extracts high-dimensional features from radiologic images, whereas pathomics applies analogous approaches to digitized histopathology, providing complementary representations of tumor biology across different spatial scales. Recent advances in foundation models and multimodal AI have accelerated the convergence of these disciplines, creating opportunities to integrate imaging, pathology, and molecular information within unified frameworks. In this review, we summarize the principles of radiomics and pathomics, discuss recent developments in foundation models for thoracic imaging and computational pathology, and examine emerging strategies for multimodal integration in lung cancer. Available evidence, although largely retrospective, suggests that integrated radio-pathomic models can outperform single-modality approaches for clinically complex tasks, including recurrence prediction, molecular characterization, and treatment response assessment. We also discuss the major barriers limiting clinical translation, including reproducibility, external validation, interoperability, interpretability, and regulatory implementation. Rather than replacing radiology or pathology, multimodal AI is progressively connecting both disciplines through shared computational representations of disease, supporting more comprehensive and biologically informed approaches to precision oncology.

Tags

Topics