18 January 2026 in medRxiv
Background: The assessment of physical examination skills in medical education is resource-intensive and prone to inter-rater variability. While artificial intelligence (AI) has successfully automated the grading of clinical notes and transcripts, evaluating the physical techniques themselves—what students do rather than what they say—remains an unsolved challenge. We evaluated whether a multimodal AI system could assess physical examination skills with expert-level reliability. Methods: In this retrospective ablation study, we analyzed 300 video-recorded encounters from six Objective Structured Clinical Examination (OSCE) stations (cardiovascular, respiratory, gastrointestinal, musculoskeletal, and neurological). We compared the performance of a multimodal AI model (Gemini 2.5 Pro) across single- and multi-camera configurations and isolated input modalities (video, audio-only, transcript-only, visual-only) against standard human grading. The primary outcome was agreement with a physician-adjudicated ground-truth reference standard, measured by quadratic weighted Cohen’s kappa (κ). Results: The AI system using a synchronized 3-camera native video configuration achieved significantly higher reliability (κ = 0.830; 95% CI, 0.773–0.880) than the standard human evaluators (κ = 0.732; 95% CI, 0.687–0.776). Performance followed a strict hierarchy: native video > audio-only > transcript-only > visual-only. Notably, visual-only models failed (κ ≈ 0.20) despite high detection accuracy, revealing a “visual paradox” where models could identify when an action occurred but not how well it was performed without audio cues. Conclusions: A properly configured multimodal AI system can grade physical examination skills with reliability exceeding that of trained human evaluators. Success requires native processing of synchronized audio-visual streams; transcript-based or visual-only approaches are insufficient for high-stakes assessment. These findings suggest that AI can provide scalable, objective, and valid assessment of clinical skills, overcoming the limitations of traditional human grading. Abstract Figure:
7 October 2026 in npj Digital Medicine
Objective Structured Clinical Examinations (OSCEs) are central to assessing medical student clinical competence, but human grading imposes substantial burden. We report, to our knowledge, the first prospective single-center deployment of an integrated multimodal artificial intelligence (AI) system for OSCE grading in undergraduate medical education, spanning notes, audio, and video. We present MAPLES (Multimodal Assessment Pipeline for Learning Encounter Scoring), a rubric-driven zero-shot multimodal LLM system deployed in Fall 2025 for 222 students (72,907 retained item-level scores). In the routed low-scoring review set, tolerant AI–standardized patient evaluator (SPE) agreement was 85.7–92.4% by modality. In a selected set of 616 disagreements adjudicated with visible score provenance, physician scores matched AI on 76.0% of items and the SPE on 19.0%. Human scoring passes fell by 92.3% versus a modeled single-pass manual comparator. Together, these findings show that multimodal AI first-pass scoring can be embedded in routine OSCE operations, with human review on the low-scoring tail and physician adjudication of selected disagreements.
21 August 2026 in Journal of Cell Biology
Submembrane cortical actin (CA) plays a large role in regulating the dynamic organization of cell surface receptors, which in turn regulates receptor signaling. Many receptors have short intracellular domains and no known link to actin. Here, we identified the β1-integrin subunit and several tetraspanins (CD9, CD81, and CD151) as part of the hitherto unknown molecular link between the receptor CD36 and CA. Our data indicate that CD36 in vascular endothelial cells interacts with these proteins, with stronger interactions near the cell edge. Compromising these interactions via the point mutation G12V in the N-terminal transmembrane domain of CD36 alters the dynamic organization of CD36 on the cell surface and weakens its coupling to CA dynamics. Moreover, it abolishes thrombospondin-1–induced CD36 signaling through the Src family kinase Fyn. Given their many interactions with transmembrane proteins, tetraspanins and integrins may provide a ubiquitous mechanism for plasma membrane–CA coupling.