Artificial intelligence for the detection and classification of occupational pneumoconioses: a systematic review

Occupational pneumoconioses remain an important occupational health problem, and chest imaging interpretation may show substantial interobserver variability. Artificial intelligence has attracted increasing interest as a supportive tool for the detection, classification, and staging of these diseases. This systematic review aimed to synthesize and critically evaluate the available evidence on the use of artificial intelligence for imaging-based assessment of occupational pneumoconioses. PubMed, Scopus, Web of Science Core Collection, and IEEE Xplore were searched for studies published from January 2015 through June 2025, in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 guidelines. Of 486 identified records, 137 duplicates were removed, 62 articles underwent full-text assessment, and 10 studies met the eligibility criteria. All studies used chest radiography, were published between 2020 and 2024, and were conducted in China and Australia. Convolutional neural networks were the predominant approach, with a shift toward ensemble models, multistage strategies, and attention mechanisms. The studies reported favorable performance but were heterogeneous in populations, diagnostic tasks, reference standards, and validation strategies. The available evidence does not allow deriving an overall estimate of diagnostic performance or establishing a consistent superiority of AI over radiologists. Artificial intelligence shows potential as a complementary tool for occupational surveillance and radiographic assessment; however, external, prospective, and multicenter validation is required before clinical and occupational implementation.