Insights on the implementation barriers of the learning methods deployment in oncology: a review
This review systematically identifies and synthesizes implementation barriers reported in the literature regarding artificial intelligence (AI)-based decision support systems in oncology. Following PRISMA guidelines, an initial search of the Scopus and PubMed databases and a subsequent citation-prioritized update identified review articles published between January 2021 and December 2025, of which 113 met the inclusion criteria. The review addresses five interconnected barriers: model interpretability, external validation, fairness and class imbalance, electronic health record integration, and data standardization. External validation is the most frequently cited challenge (73.5%), followed by model interpretability (66.4%) and data standardization (65.5%). Temporal trends show a marked increase in validation concerns, from 56.7% in 2021 to 93.8% in 2025, and declining attention to fairness barriers, from 53.3% in 2021 to 37.5% in 2025. This paper makes four main contributions. First, it presents an integrative framework showing that the barriers form an interconnected system in which fairness propagates into all downstream components, so that the observed decline in explicit fairness research signals a critical gap in current priorities. Second, it quantifies barrier prevalence and its temporal evolution, enabling evidence-based prioritization. Third, it consolidates the approaches reported in the literature into unified pipelines for bias mitigation and model interpretation. Fourth, it proposes a stage-specific implementation protocol with a decision flowchart to guide barrier assessment before deployment in screening, diagnosis, treatment, and prognosis applications. The findings indicate that effective AI implementation in oncology requires addressing multiple barriers simultaneously rather than in isolation. The proposed frameworks and implementation protocol provide actionable guidance for researchers and practitioners working to translate AI models from development to clinical practice.
Authors
- Alexander Caicedo (ORCID: https://orcid.org/0000-0002-2073-8008)
- Carlos Ocampo‐Martínez (ORCID: https://orcid.org/0000-0001-9251-6044)
- Johan Pena-Campos
- Diego Patiño
Institutions
- Pontificia Universidad Javeriana (CO)
- Universitat Politècnica de Catalunya (ES)
Publication Details
- Journal
- Frontiers in Digital Health
- Published
- 2026-11-19
- DOI
- https://doi.org/10.3389/fdgth.2026.1813475
- Primary Topic
- Artificial Intelligence in Healthcare and Education
- Type
- article
- Field-Weighted Citation Impact
- 0.00