fig3

Figure 3. From model performance to clinical benefit: where current evidence stops. Counts show how many of the 31 direct AI studies attained each cumulative evidence level. The greatest gap lay between validation of performance and clinical actionability: 13 studies attained at least Level 2, only three attained Level 3, and none prospectively assessed decision impact or patient outcomes. Source: Original diagram prepared for this review from the evidence synthesis and evidence-maturity framework. AI: Artificial intelligence.