Table 3 Performance comparison of unimodal and LLMMs in new-onset T2DM prediction using different modality combinations of A, B, and C.

Architecture	Modality	Model	Accuracy	Recall	Precision	F1-score
Machine Learning Classifier	C	Logistc Regeression	0.79	0.79	0.79	0.73
		KNN	0.83	0.83	0.81	0.81
		K-Means	0.83	0.83	0.82	0.81
		SVM	0.78	0.78	0.80	0.71
		Random Forest	0.86	0.86	0.85	0.85
		XGboost	0.86	0.86	0.85	0.85
		CatBoost	0.86	0.86	0.85	0.85
		DNN (Three-layer)	0.85	0.53	0.74	0.61
Unimodal	A	BiomedBERT	0.65	0.65	0.66	0.65
		ClinicalBERT	0.61	0.61	0.61	0.61
		SciFive	0.66	0.66	0.66	0.66
		RoBERTa	0.65	0.65	0.65	0.65
		Flan-T5-base-220M	0.62	0.62	0.67	0.62
		Flan-T5-large-770M	0.79	0.79	0.78	0.79
		BERT	0.66	0.66	0.66	0.66
		GPT-2	0.78	0.78	0.77	0.77
Unimodal	A+B	BiomedBERT	0.82	0.82	0.82	0.82
		ClinicalBERT	0.78	0.78	0.79	0.77
		SciFive	0.76	0.76	0.76	0.76
		RoBERTa	0.83	0.83	0.83	0.83
		Flan-T5-base-220M	0.81	0.81	0.82	0.81
		Flan-T5-large-770M	0.93	0.93	0.93	0.93
		BERT	0.79	0.79	0.79	0.79
		GPT-2	0.93	0.93	0.93	0.93
LLMMs	A+C	BiomedBERT	0.81	0.81	0.80	0.80
		ClinicalBERT	0.81	0.81	0.80	0.80
		SciFive	0.81	0.81	0.80	0.80
		RoBERTa	0.80	0.80	0.79	0.80
		Flan-T5-base-220M	0.80	0.80	0.80	0.80
		Flan-T5-large-770M	0.81	0.81	0.81	0.81
		BERT	0.82	0.82	0.80	0.81
		GPT-2	0.81	0.81	0.79	0.81
LLMMs	(A+B) + C	BiomedBERT	0.93	0.93	0.93	0.93
		ClinicalBERT	0.90	0.90	0.90	0.91
		SciFive	0.93	0.93	0.93	0.93
		RoBERTa	0.92	0.92	0.92	0.92
		Flan-T5-base-220M	0.93	0.93	0.93	0.93
		Flan-T5-large-770M	0.92	0.92	0.93	0.92
		BERT	0.93	0.93	0.93	0.93
		GPT-2	0.92	0.92	0.92	0.92

Quick links

Search