Table 4.

Overview of LMs for captioning medical images. The asterisks (*) indicate terms that are either not present in the original paper or do not apply in this context.

References	ROI	Modality	Dataset	Model name	Vision model	Language model
Nicolson et al (2023)	*	x-ray, Ultrasound, CT, MRI	Radiology objects in context (ROCO)	CvT2DistilGPT2	CvT-21	DistilGPT2
Nicolson et al (2021)	*	CT, Ultrasoud, x-ray, Fluroscopy, PET, Mammography, MRI, Angiography	ROCO, ImageCLEFmed Caption 2021	*	ViT	PubMedBERT
Kim et al (2023b)	Brain	CT	Institutional	*	ResNet-50, EfficientNet-B5, DenseNet-201, and ConvNeXt-S	DistilGPT2
Zheng and Yu (2023)	*	*	IU-Xray	*	CDGPT (Conditioned Densely-connected Graph Transformer)	AlignTrans (Alignment Transformer)
Wang and Li (2022)	*	*	ImageCLEFmedical Caption 2022	BLIP	Vision Transformer (ViT-B)	BERT
Ding et al (2023)	*	Histopathology	MIDOG 2022	CLIP, BLIP	Vision Transformer (ViT-B)	BERT
Wang et al (2023a)	*	*	ImageCLEFmedical Caption 2022	BLIP	Vision Transformer (ViT-B)	BERT
Zhou et al (2023)		Ultrasound, CT, x-ray, MRI	ROCO	BLIP-2	ViT-g/14	OPT2.7B