Modality-Aware Trustworthy Data Access for Multimodal Healthcare Fusion in IoMT-Enabled MLLMs

The Internet of Medical Things (IoMT) generates distributed multimodal health data that support multimodal large language models (MLLMs) for diagnosis, risk prediction, and clinical decision support. Trustworthy healthcare inference requires fine-grained control over modality access under task-specific authorization and privacy constraints. We propose a modality-aware trustworthy data access framework that independently authorizes requested modalities and enforces task-specific access policies. By integrating BBS+ signatures, zero-knowledge proofs, and authenticated key agreement, the framework supports selective disclosure of certified requester attributes while protecting unrelated information and long-term identity. Credential presentations are bound to the requested modality, healthcare task, revocation state, and accountability context, enabling auditable access decisions. Security analysis establishes modality authorization, mutual authentication, session-key security, selective-disclosure privacy, revocation enforcement, and controlled accountability. Evaluation on the OpenI chest X-ray dataset with MedGemma demonstrates how authorized modality availability influences downstream inference, with joint image–text input outperforming image-only inference. The framework further achieves millisecond-level local cryptographic processing under a constrained gateway profile and supports authorization over IEEE 802.15.4/6LoWPAN with scalable revocation synchronization. These results demonstrate the feasibility of privacy-preserving modality access governance for multimodal healthcare inference.