The progress of smart healthcare based on Multimodal Data fusion technology
TL;DR
This paper analyzes the challenges of existing methods in terms of data heterogeneity, modal absence, interpretability, interpretability, and privacy protection, explores the deficiencies in the dataset construction and evaluation system, and proposes corresponding solutions.
Abstract
Smart healthcare is moving towards precision, personalization, and intelligence. Multimodal data fusion technology, as a core means to break medical data silos and improve clinical decision-making efficiency, has received extensive attention. This paper systematically reviews the mainstream methods and their performance in typical medical tasks from four levels: data level, feature level, decision level, and hybrid fusion. Based on this, the current commonly used medical multimodal datasets and evaluation criteria are introduced, and the experimental results of various fusion methods on public datasets are summarized. Further, this paper analyzes the challenges of existing methods in terms of data heterogeneity, modal absence, interpretability, and privacy protection, explores the deficiencies in the dataset construction and evaluation system, and proposes corresponding solutions. Finally, it looks forward to future research directions, emphasizing the synergy of medical-engineering integration, privacy computing and explainable models to provide references for the in-depth research and application of multimodal fusion technologies in smart healthcare.