Clinical bottom line
Radiation-induced pulmonary fibrosis (RIPF) is a serious late complication of thoracic radiotherapy that can significantly impair respiratory function. Early detection is critical for timely intervention and management, but conventional imaging often identifies fibrosis only after irreversible damage has occurred. Emerging evidence suggests that AI-enhanced imaging techniques, including radiomics and deep learning applied to CT scans, may improve early detection of RIPF by identifying subtle imaging features before clinical symptoms or overt radiographic changes develop. However, these approaches remain largely investigational and require further validation in diverse clinical settings.
What the evidence shows
Radiation-induced pulmonary fibrosis results from complex tissue remodeling following radiation injury, characterized by fibroblast proliferation and extracellular matrix deposition (Schubert et al., 2012; Bickle et al., 2020). Conventional imaging modalities such as high-resolution CT (HRCT) are the current standard for diagnosing established fibrosis but have limited sensitivity for early changes (Adame et al., 2015).
Recent advances in AI-enhanced imaging leverage radiomics — the extraction of quantitative imaging features — combined with deep learning algorithms to detect subtle textural and structural alterations in lung parenchyma that precede visible fibrosis (Ivković et al., 2026). Ivković et al. (2026) reported a multicenter study demonstrating that radiomics-based models could differentiate early RIPF from normal post-radiation changes with promising accuracy, outperforming standard radiologic assessment. Similarly, Kiziloğlu et al. (2026) showed that attention mechanism-enhanced deep learning models improved differential diagnosis of fibrotic lung diseases on CT, suggesting potential applicability to RIPF.
These AI approaches analyze complex imaging patterns beyond human visual perception, potentially enabling earlier detection and risk stratification. However, most studies to date are retrospective, with relatively small cohorts, and have not yet been integrated into routine clinical workflows.
Caveats and uncertainty
While AI-enhanced imaging for early RIPF detection is promising, several limitations warrant caution. The heterogeneity of radiation protocols, timing of imaging post-therapy, and patient comorbidities complicate model generalizability (Qureshi et al., 2018). Many AI models require large, well-annotated datasets for training and validation, which remain scarce for RIPF.
Moreover, distinguishing early fibrosis from radiation pneumonitis or other inflammatory changes remains challenging, even for AI algorithms. The clinical significance of AI-detected subtle imaging abnormalities is not fully established, and their predictive value for progression or response to therapy is unclear.
Finally, integration of AI tools into clinical practice demands regulatory approval, standardization, and clinician training, all of which are ongoing hurdles.
How this may change practice
If validated in prospective studies, AI-enhanced imaging could transform RIPF management by enabling earlier diagnosis, personalized risk assessment, and timely therapeutic interventions. Radiologists may increasingly incorporate AI-driven quantitative imaging analyses into routine post-radiation surveillance protocols, complementing clinical and pulmonary function assessments.
Early identification of patients at high risk for fibrosis could facilitate enrollment in clinical trials of antifibrotic agents or other novel therapies. Furthermore, AI tools might assist radiation oncologists in tailoring radiation plans to minimize pulmonary toxicity.
However, until robust evidence and clinical guidelines emerge, AI-enhanced imaging for RIPF should be considered investigational and used adjunctively rather than as a standalone diagnostic tool.