notesum.ai

Published at October 20

Concept Complement Bottleneck Model for Interpretable Medical Image Diagnosis

cs.CV

cs.AI

Released Date: October 20, 2024

Authors: Hongmei Wang¹, Junlin Hou¹, Hao Chen¹

Aff.: ¹Not specified

Arxiv: https://arxiv.org/abs/2410.15446v1

Dataset	Model	Disease Diagnosis			Concept Detection
Dataset	Model	AUC (%) $\uparrow$	ACC (%) $\uparrow$	F1-score (%) $\uparrow$	AUC (%) $\uparrow$	ACC (%) $\uparrow$
Derm7pt	PCBM [30]	81.32 $\pm$ 2.12	75.82 $\pm$ 2.00	71.10 $\pm$ 1.74	N/A	N/A
	PCBM-H [30]	85.87 $\pm$ 1.53	78.60 $\pm$ 2.79	75.50 $\pm$ 3.09	N/A	N/A
	CBIVLM [21]	83.45 $\pm$ 3.59	84.13 $\pm$ 2.71	71.24 $\pm$ 2.54	N/A	N/A
	ECBM [33]	77.44 $\pm$ 1.84	79.69 $\pm$ 1.84	76.57 $\pm$ 0.00	59.54 $\pm$ 1.79	77.73 $\pm$ 1.47
	AEC [34]	91.27 $\pm$ 2.02	84.88 $\pm$ 2.05	80.99 $\pm$ 3.32	76.61 $\pm$ 1.61	75.30 $\pm$ 0.72
	CBM [24]	92.88 $\pm$ 1.90	85.89 $\pm$ 1.92	82.18 $\pm$ 2.67	82.15 $\pm$ 2.68	80.00 $\pm$ 1.87
	CCBM	93.96 $\pm$ 0.95	88.15 $\pm$ 1.64	85.61 $\pm$ 2.79	83.86 $\pm$ 1.00	80.90 $\pm$ 1.56
	Inception-v3	92.02 $\pm$ 2.53	86.46 $\pm$ 2.54	83.13 $\pm$ 3.31	N/A	N/A
Skincon	PCBM [30]	69.06 $\pm$ 1.23	72.48 $\pm$ 1.56	39.55 $\pm$ 0.97	N/A	N/A
	PCBM-H [30]	72.85 $\pm$ 1.66	68.42 $\pm$ 3.07	53.33 $\pm$ 3.38	N/A	N/A
	CBIVLM [21]	71.34 $\pm$ 2.16	70.74 $\pm$ 1.11	66.15 $\pm$ 1.50	N/A	N/A
	ECBM [33]	68.72 $\pm$ 1.73	78.79 $\pm$ 1.41	61.12 $\pm$ 0.00	65.11 $\pm$ 0.89	90.94 $\pm$ 0.25
	AEC [34]	83.86 $\pm$ 0.61	80.24 $\pm$ 0.52	63.89 $\pm$ 1.63	58.64 $\pm$ 0.90	90.64 $\pm$ 0.10
	CBM [24]	80.01 $\pm$ 1.25	78.42 $\pm$ 1.31	60.57 $\pm$ 3.04	62.14 $\pm$ 1.37	89.32 $\pm$ 0.14
	CCBM	84.55 $\pm$ 1.87	80.00 $\pm$ 1.75	67.14 $\pm$ 1.87	82.14 $\pm$ 0.33	91.35 $\pm$ 0.25
	Inception-v3	79.92 $\pm$ 1.48	77.52 $\pm$ 1.47	59.86 $\pm$ 2.78	N/A	N/A
BrEaST	PCBM [30]	75.41 $\pm$ 5.74	68.43 $\pm$ 7.64	64.79 $\pm$ 8.28	N/A	N/A
	PCBM-H [30]	79.63 $\pm$ 2.95	70.02 $\pm$ 3.87	67.73 $\pm$ 3.81	N/A	N/A
	ECBM [33]	71.38 $\pm$ 3.84	73.63 $\pm$ 5.28	71.33 $\pm$ 0.00	62.50 $\pm$ 2.67	79.94 $\pm$ 1.79
	AEC [34]	82.80 $\pm$ 3.32	72.40 $\pm$ 3.39	68.60 $\pm$ 3.51	69.71 $\pm$ 1.52	77.82 $\pm$ 1.70
	CBM [24]	87.42 $\pm$ 4.27	77.21 $\pm$ 8.62	76.29 $\pm$ 8.31	70.76 $\pm$ 1.38	77.49 $\pm$ 1.43
	CCBM	88.49 $\pm$ 5.85	79.21 $\pm$ 6.00	77.95 $\pm$ 5.53	77.78 $\pm$ 4.00	80.06 $\pm$ 3.46
	ResNet50	86.97 $\pm$ 6.14	77.61 $\pm$ 6.23	76.39 $\pm$ 6.09	N/A	N/A
		AUC (%) $\uparrow$	ACC (%) $\uparrow$	F1-score (%) $\uparrow$	RMSE $\downarrow$	MAE $\downarrow$
LIDC-IDRI	PCBM [30]	85.09 $\pm$ 1.33	81.47 $\pm$ 1.24	77.96 $\pm$ 1.59	N/A	N/A
	PCBM-H [30]	88.77 $\pm$ 1.64	83.28 $\pm$ 1.57	80.71 $\pm$ 1.46	N/A	N/A
	AEC [34]	87.35 $\pm$ 2.16	84.23 $\pm$ 1.67	81.57 $\pm$ 1.86	0.5710 $\pm$ 0.0100	0.5064 $\pm$ 0.0094
	CBM [24]	89.76 $\pm$ 1.30	83.63 $\pm$ 1.67	80.22 $\pm$ 1.85	0.2981 $\pm$ 0.0096	0.2461 $\pm$ 0.0095
	CCBM	90.48 $\pm$ 0.94	84.82 $\pm$ 0.43	82.56 $\pm$ 0.67	0.1890 $\pm$ 0.0046	0.1380 $\pm$ 0.0032
	ResNet50	90.18 $\pm$ 1.17	84.08 $\pm$ 1.71	81.70 $\pm$ 1.52	N/A	N/A