Artificial intelligence and radiomics in drug-induced interstitial lung disease.

Background: Drug-induced interstitial lung disease (DI-ILD) encompasses a heterogeneous spectrum of potentially severe pulmonary toxicities associated with various pharmacological agents. Diagnosis is often delayed due to nonspecific symptoms and imaging findings that can mimic other ILDs or infections, particularly in patients receiving oncological or immunomodulatory therapies. Novel approaches are needed to improve early recognition and management. Methods: We conducted a non-systematic, narrative literature review across major biomedical databases up to June 2025, aimed at describing current and emerging applications of artificial intelligence (AI) and radiomics in the early diagnosis, risk stratification, and personalised management of DI-ILD. Results: Radiomics applied to computed tomography and positron emission tomography/computed tomography enables extraction of high-dimensional quantitative features capturing subclinical alterations undetectable by visual assessment. In DI-ILD (particularly immune checkpoint inhibitor-related pneumonitis), radiomic models show potential diagnostic utility in distinguishing overlapping imaging patterns and predicting fibrotic progression. Integration with clinical, radiological and pharmacogenomic data has improved model performance in several studies. Additionally, AI approaches, including convolutional neural networks and ensemble learning methods, demonstrate promise in enhancing pattern recognition and risk stratification. Conclusions: Radiomics and AI are emerging complementary tools in multidisciplinary management of DI-ILD, offering objective imaging biomarkers and facilitating multimodal data integration to improve diagnostic precision and personalised therapeutic decisions. Nonetheless, reproducibility remains limited by variability in imaging protocols and lack of large-scale prospective multicentre validation. Clinical implementation requires standardised protocols to ensure consistency and reliability. The development of transparent, interpretable models seamlessly integrated into healthcare workflows is essential to fully leverage their real-world potential in proactive patient care.