A three-stage maturity model for clinical translation of pathology AI (IMAGE)
Caption
A three-stage maturity model for clinical translation of pathology AI, linking structural barriers to system level pathways. The model organizes the translational journey from research prototype to sustained clinical use into three sequential stages, each gated by specific barriers (analyzed in Section 3) and addressed by corresponding pathways (Section 4). Stage 1: Algorithmic capability (gated by retrospective benchmark evidence). Barrier: data and infrastructure fragility (scanner shift, format fragmentation, manual QC). Pathway: infrastructure first AI (vendor-neutral formats, automated QC, multicenter baselines, federated learning). Stage 2: System integration (gated by prospective multisite validation). Barrier: workflow and human-system misalignment (cognitive rhythm, automation bias, scenario-dependent latency). Pathway: workflow embedded intelligence (triage, assistive layers, uncertainty visualization, frictionless human override). Stage 3: Institutional adoption (gated by demonstrable workflow benefit, viable reimbursement, and a documented governance plan). Barrier: institutional trust and governance constraints (interpretability, validation gaps, liability, generative AI risks). Pathway: adaptive governance [machine learning operations (MLOps), shadow deployment, real-world evidence, PCCPs for continuous learning]. The bottom row maps representative approved and research products (from Table 1) to their current dominant maturity stage as of early 2026. This framework enables systematic diagnosis of why a given AI system fails to reach clinical practice and what intervention is required to advance it to the next stage.
Credit
Lu Cai, Biwen Meng, Jie Huang, Guanyu Ding, Min Ju, Wenwen Wang, Shijie Deng, Liqin Lai, Jin Wang, Chunxue Yang, Miao Ruan, Shugong Xu, Chaofu Wang, Jingxin Liu, Qian Da.
Usage Restrictions
Credit must be given to the creator. Only noncommercial uses of the work are permitted. No derivatives or adaptations of the work are permitted.
License
CC BY-NC-ND