Table 13.

Comparative assessment of limitations and future perspectives for research in multimodal learning.

References	Specific limitation	Specific future perspective
Huemann et al (2023a)	Single dataset; limited multimodal datasets; primary annotator bias; language as input requirement; comparison challenges	Diverse datasets; inter-observer variability; task expansion; reducing language dependency; AI-assisted clinical workflow
Huang et al (2021)	Focus on chest radiographs; dependence on report quality; designed for english-language reports	Expand to other modalities and regions; improve robustness to report quality; adapt for different languages
Li et al (2023d)	2D segmentation limitation; manual text annotation requirement	3D segmentation; automating text annotation generation
Khare et al (2021)	Small labeled datasets; model interpretability; incomplete consideration of expert diagnosis	Larger datasets; enhanced interpretability; clinical integration
Chen et al (2022)	Complexity and diversity handling in medical image-text data	Sophisticated techniques for diverse medical scenarios
Delbrouck et al (2022)	Visio-linguistic reasoning and understanding; overfitting and generalization; transparency in medical AI; NLG evaluation in medical AI	Enhanced interpretability in multimodal scenarios; refined training methodologies; improved documentation and accessibility; advanced evaluation metrics