Table 2.

Performance of few-shot learning models and radiologists in Cohort #1, round #2.

Round #2	Confusion Matrix					Metrics
		VGG16+XGBoost
		Normal	OP	OS	Total	Accuracy	0.688
Actual	Normal	39	11	29	79	Sensitivity (OP)	0.648
	OP	6	59	26	91	Sensitivity (OS)	0.857
	OS	5	11	96	112	Sensitivity (mean)	0.753
	Total	50	81	151	282	Specificity	0.494
		ResNet50+XGBoost
		Normal	OP	OS	Total	Accuracy	0.585
Actual	Normal	22	24	33	79	Sensitivity (OP)	0.670
	OP	5	61	25	91	Sensitivity (OS)	0.732
	OS	4	26	82	112	Sensitivity (mean)	0.701
	Total	31	111	140	282	Specificity	0.278
		Xception+XGBoost
		Normal	OP	OS	Total	Accuracy	0.709
Actual	Normal	36	18	25	79	Sensitivity (OP)	0.725
	OP	3	66	22	91	Sensitivity (OS)	0.875
	OS	5	9	98	112	Sensitivity (mean)	0.800
	Total	44	93	145	282	Specificity	0.456
		Junior radiologist
		Normal	OP	OS	Total	Accuracy	0.436
Actual	Normal	47	19	13	79	Sensitivity (OP)	0.330
	OP	30	30	31	91	Sensitivity (OS)	0.411
	OS	26	40	46	112	Sensitivity (mean)	0.370
	Total	103	89	90	282	Specificity	0.595
		Senior radiologist
		Normal	OP	OS	Total	Accuracy	0.504
Actual	Normal	52	16	11	79	Sensitivity (OP)	0.363
	OP	33	33	25	91	Sensitivity (OS)	0.509
	OS	14	41	57	112	Sensitivity (mean)	0.436
	Total	99	90	93	282	Specificity	0.658

Bold figures indicate the highest numeric values.

OP, osteopenia; OS, osteoporosis; XGBoost, eXtreme gradient boosting.