Rebuilding ability acquisition in medical imaging teaching: a multi-level AI-driven teaching paradigm based on cognitive-information theory

Medical imaging education faces challenges like learner variability and delayed feedback, hindering efficiency and skill transfer. While AI shows promise, no previous studies have systematically tested multi-layer AI teaching frameworks using dual-level randomized experiments. This study evaluates a cognitive informatics–driven “three-level–four-module” AI teaching paradigm. Sixty postgraduate imaging students participated in a course-level RCT (AI paradigm vs. traditional teaching, 8–12 weeks) and a 2 × 2 factorial experiment to dissect the effects of “explainable guidance” and “adaptive scheduling.” Data were analyzed using linear mixed-effects models and DeLong tests. The AI paradigm significantly improved diagnostic performance (AUC: β = 0.072, P < 0.001; accuracy: +5.8%, P = 0.004) and reduced cognitive load (NASA-TLX, η²=0.12, P < 0.01). Factorial analysis revealed both components independently increased AUC (Δ ≈ 0.05, P < 0.01) with a positive synergistic interaction (β = 0.021, P = 0.03), and cognitive load reduction was maximal when combined (Δ=–17.6, P = 0.01). The AI-driven paradigm enhances learning effectiveness and reduces cognitive load, addressing a critical gap by providing scalable, evidence for multi-layer AI integration in medical imaging instruction.