AI-Based Post-processing for Artefact Mitigation in Radiography: A Systematic Review.
Authors
Affiliations (6)
Affiliations (6)
- Monash Radiology, Monash Health, 246 Clayton Rd, Clayton, Melbourne, VIC, 3168, Australia. [email protected].
- Department of Medical Imaging and Radiation Sciences, School of Primary and Allied Health Care, Faculty of Medicine, Nursing and Health Sciences, Monash University, Melbourne, Australia. [email protected].
- Medical Imaging and Interventional Radiology Department, Fiona Stanley and Fremantle Hospitals Group, Perth, Australia.
- Library Services, Monash Health, Melbourne, Australia.
- Department of Medical Imaging and Radiation Sciences, School of Primary and Allied Health Care, Faculty of Medicine, Nursing and Health Sciences, Monash University, Melbourne, Australia.
- Monash Radiology, Monash Health, 246 Clayton Rd, Clayton, Melbourne, VIC, 3168, Australia.
Abstract
Projectional radiography is vulnerable to artefacts that can impair image quality and obscure or mimic pathology, confounding image interpretation. This systematic review synthesises published research on AI-based post-processing methods for artefact mitigation in radiography, identifies challenges to clinical translation, and evaluates reporting practices against the Checklist for Artificial Intelligence in Medical Imaging (CLAIM). Literature searches were performed across multiple bibliographic databases for studies published before 09/01/2026. A broad radiology screening strategy was used, after which studies specific to radiography were selected. Original peer-reviewed articles on AI-based post-processing methods for artefact mitigation in radiography were included. Exclusion criteria that included AI methods requiring access to pre-reconstruction data, hardware-based approaches, did not involve human subjects, and non-English languages publications. Screening was performed independently by two reviewers, and included studies were assessed against CLAIM. The review protocol was registered with PROSPERO (CRD420251089690). The search identified 2965 records, of which 10 studies were included. Four application categories were identified: bone shadow removal in chest X-rays, immobilisation device removal, electronic device removal, and general acquisition artefact removal. Studies were published between 2020 and 2025, with generative adversarial networks (GANs) the predominant architecture. No study was prospective, and methodological details were typically not sufficiently reported to allow reproducibility. CLAIM adherence was limited, with a mean score of 17.4 ± 3.6 out of 42 applicable items. The current evidence base is insufficient to support clinical adoption. Future work should prioritise external multi-institutional testing, clinically meaningful endpoints, failure mode analysis, open data and code, and workflow-embedded, prospective study design.