Undergraduate programs in intelligent medical engineering are expanding, yet laboratory curricula lag behind the multimodal, long‑tailed, and distribution‑shifting realities of clinical AI. This paper translates an ongoing multimodal deep‑learning project on endometrial carcinoma into a structured experimental teaching system. Three educational gaps are identified: insufficient modality coverage, lack of authentic data, and missing deployment practice. From constructive alignment, experiential learning, the research‑teaching nexus, and the CDIO framework, four pedagogical principles are derived. The curriculum consists of four progressive tiers plus an engineering layer, totaling 32 laboratory units and 64 contact hours. All labs run on a custom virtual clinical workstation using de‑identified multi‑institutional datasets. Each tier targets a specific technical bottleneck, prerequisite coursework, and criterion‑referenced deliverable, such as data preprocessing, modality alignment, cross‑domain adaptation, and model deployment. Data governance, safety, and assessment protocols are defined, and learning outcome data will be gathered across two implementation cycles.
Review: This system bridges cutting‑edge research and hands‑on teaching, addressing modality diversity, data authenticity, and deployment skills, thereby offering a comprehensive pathway for training future medical AI professionals.