Diffusion tensor imaging for characterizing tumor microstructure and improving diagnostic performance on breast MRI: a prospective observational study

Table 5 Multivariate Clinical/DCE-MRI+DTI LASSO model with type- and size-specific DTI parameters

	Odds ratios*
	Model without interactions	Model with interactions^†
Clinical/DCE-MRI parameters
Age, per 1-SD increase	1.17	1.20
Post-menopausal	NS	NS
MRI indication: known cancer	1.88	1.84
Dense breasts	0.61	0.50
BPE category, per 1-category increase	0.66	0.59
Lesion size^‡, per 1-SD increase	1.85	2.18
Mass vs. NMLE/focus	1.89	2.28
Washout on delayed phase kinetics	2.43	3.07
BI-RADS 5 vs. 4	2.36	2.97
DTI parameters
Mean ADC, per 1-SD increase
Small non-masses	0.41	1.00
Large non-masses	0.41	0.35
Small masses	0.41	0.36
Large masses	0.41	0.13
Mean FA^‡, per 1-SD increase
Small non-masses	0.88	0.47
Large non-masses	0.88	0.41
Small masses	0.88	1.17
Large masses	0.88	1.02
Bootstrap-adjusted AUC	0.81	0.85
(95% CI)	(0.78, 0.88)	(0.82, 0.90)

Small lesion, < 1 cm; large lesion, ≥ 1 cm
*For continuous variables, ORs are scaled to show change per 1-SD increase in the corresponding variable
^†The model with interactions included addition terms corresponding to type × ADC, size × ADC, type × FA, and size × ADC, allowing all 4 subgroups (type × size) to have different ORs for ADC and different ORs for FA; the model without interactions contains 12 regression parameters (including the intercept), and the model with interactions contains 16 regression parameters
^‡Variable was log-transformed prior to inclusion in the logistic regression model to reduce right-skewness
NS = variable was included as a candidate predictor but was not selected by the LASSO

ISSN: 1465-542X