Identifying factors that may influence the classification performance of radiomics models using contrast-enhanced mammography (CEM) images

Table 2 Summary of image features and objective quantitative features in subgroups of interest

Image features	Category	Benign lesions		P value	Malignant lesions		P value
Image features	Category	High Misclassification Probability (n = 5)	Low Misclassification Probability (n = 32)	P value	High Misclassification Probability (n = 11)	Low Misclassification Probability (n = 69)	P value
Lesion size ^a	/	31.3 ± 11.8	11.8 ± 9.2	0.003	17.0 ± 6.0	34.4 ± 16.8	< 0.001
Breast density	a-b	1/5 (40.0)	2/32 (6.3)	0.362	2/11 (18.2)	23/69 (33.3)	0.488
	c-d	4/5 (60.0)	30/32 (93.8)		9/11 (81.8)	46/69 (66.7)
Degree of BPE	Minimal or mild	1/5 (20.0)	14/32 (43.8)	0.629	10/11 (90.9)	52/69 (75.4)	0.440
	Moderate or marked	4/5 (80.0)	18/32 (56.3)	0.629	1/11 (9.1)	17/69 (24.6)	0.440
Rim artifact	Absent	1/5 (20.0)	31/32 (96.9)	< 0.001	10/11 (90.9)	59/69 (85.5)	0.999
	Present	4/5 (80.0)	1/32 (3.1)	< 0.001	1/11 (9.1)	10/69 (14.5)	0.999
Ripple artifact	Absent	3/5 (60.0)	31/32 (96.9)	0.042	6/11 (54.5)	33/69 (47.8)	0.753
	Present	2/5 (40.0)	1/32 (3.1)	0.042	5/11 (45.5)	36/69 (52.2)	0.753
Vascular artifact	Absent	4/5 (80.0)	29/32 (90.6)	0.456	9/11 (81.8)	48/69 (69.6)	0.497
	Present	1/5 (20.0)	3/32 (9.4)	0.456	2/11 (18.2)	21/69 (30.4)	0.497
Air trapping artifact	Absent	5/5 (100.0)	31/32 (96.9)	0.255	7/11 (63.6)	57/69 (82.6)	0.217
	Present	0/5 (0.0)	1/32 (3.1)	0.255	4/11 (36.4)	12/69 (17.4)	0.217
SNR ^a	/	467.6 ± 100.8	143.9 ± 87.3	< 0.001	224.1 ± 75.9	410.4 ± 130.0	< 0.001
CNR ^a	/	539.7 ± 73.8	149.5 ± 98.4	< 0.001	247.3 ± 92.1	505.2 ± 179.9	< 0.001
BCR ^a	/	299.8 ± 38.4	83.8 ± 53.4	< 0.001	133.6 ± 49.7	265.7 ± 93.8	< 0.001

BPE background parenchymal enhancement, SNR signal-to-noise ratio, CNR contrast-to-noise ratio, BCR background contrast ratio
^a Data are shown as the mean values ± standard deviations. Other data are shown as proportions with percentages in parentheses

ISSN: 1470-7330