Radiomics in clinical radiology: advances, challenges, and future directions

Abstract

High-throughput extraction of quantitative image features, known as radiomics, has the potential to improve clinical radiology by revealing hidden features in medical images. In this review, we discuss the convergence of radiomics with artificial intelligence (AI), specifically the development of large language models (LLMs) and agentic AI models, alongside improvements in radiomics methods, standardisation, and clinical uptake pathways. Our aim is to summarise these developments, highlight current challenges, and suggest potential future directions for the widespread adoption of radiomics in clinical practice. The study conducted an extensive literature review, focusing on radiomics research, particularly investigations that examined validation frameworks, standardisation efforts, deep learning, LLMs, multi-centre studies, and the emerging library of agentic pipelines. The review compares key publications to outline common findings affecting diagnostic accuracy, prognostic performance, reproducibility, and integration into clinical workflows. Advances in AI, particularly LLMs that work with both text and images to enhance understanding, and agentic AI systems that tailor workflows to specific conditions, offer promising opportunities to improve radiomic analysis and its clinical applicability. Issues of reproducibility are being addressed through standardisation and robust validation, paving the way for broader clinical implementation. Radiomics has the potential to become an essential component of precision imaging in the coming years, though this will require resolving challenges related to reproducibility, interpretability, and seamless integration into clinical practice through cross-disciplinary collaboration. CLINICAL RELEVANCE: Radiologists and imaging scientists should anticipate the future use of radiomic AI solutions, potentially incorporating LLMs, to support diagnostic, prognostic, and therapeutic decision-making, ultimately enhancing precision medicine.

Ähnliche Arbeiten