Computational and AI-Enabled Imaging Biomarkers for Predicting and Assessing Immunotherapy Response in Oral Squamous Cell Carcinoma: A Systematic Review with Functional Meta-Synthesis.
Authors
Affiliations (3)
Affiliations (3)
- Department of Periodontics, Saveetha Institute of Medical and Technical Sciences, Saveetha Dental College and Hospitals, Saveetha University, Chennai 600077, India.
- Biomedical Stomatology Research Group, Basic Sciences Department, Faculty of Dentistry, Universidad de Antioquia U de A, Medellín 050010, Colombia.
- Faculty of Dentistry, Institución Universitaria Visión de las Américas, Medellín 050040, Colombia.
Abstract
<b>Background/Objectives:</b> Computational and artificial intelligence (AI)-enabled imaging biomarkers are increasingly being investigated for immunotherapy-response assessment in oral squamous cell carcinoma (OSCC), but the evidence is heterogeneous. This systematic review identified, critically appraised, and synthesized imaging biomarkers used to predict or assess observed immunotherapy response in OSCC. <b>Methods:</b> PubMed/MEDLINE, Embase, and Scopus were searched from inception to 31 July 2026, without language restrictions. The protocol was registered in PROSPERO. Eligible studies evaluated quantitative, radiomic, machine-learning, or related medical-imaging biomarkers against pathological or radiological response. Risk of bias was assessed using design-specific tools, supplemented by a radiomics-specific methodological appraisal. A structured Synthesis Without Meta-analysis was prespecified. <b>Results:</b> Seven reports were included, covering pretreatment prediction, early/on-treatment dynamic assessment, and post-neoadjuvant/preoperative response assessment. Pretreatment-computed tomography and magnetic resonance imaging radiomics showed discriminatory performance, whereas baseline positron emission tomography uptake alone was not consistently associated with pathological response. Longitudinal fibroblast activation protein inhibitor positron emission tomography changes were associated with a major pathological response, and magnetic resonance imaging habitat and longitudinal habitat models showed higher within-study discrimination when combined with clinical information. No computational model had independent external validation specifically in an OSCC population, and no three independent studies were sufficiently comparable for meta-analysis. <b>Conclusions:</b> Computational imaging is promising across several stages of immunotherapy-response assessment. However, based on current evidence, none of the reviewed imaging biomarkers or models can yet be considered ready for routine clinical use. Prospective multicenter OSCC-specific validation and standardized imaging and analytical pipelines are needed to establish generalizability and clinical utility.