Untitled

Funder

Organizational Unit

Publications

Improving malignancy prediction in breast lesions with the combination of apparent diffusion coefficient and dynamic contrast-enhanced kinetic descriptors

Publication . Nogueira, Luisa; Brandão, Sofia; Matos, Eduarda; Gouveia Nunes, Rita; Ferreira, Hugo Alexandre; Loureiro, Joana; Ramos, Isabel

Aim To assess how the joint use of apparent diffusion coefficient (ADC) and kinetic parameters (uptake phase and delayed enhancement characteristics) from dynamic contrast-enhanced (DCE) can boost the ability to predict breast lesion malignancy. Materials and methods Breast magnetic resonance examinations including DCE and diffusion-weighted imaging (DWI) were performed on 51 women. The association between kinetic parameters and ADC were evaluated and compared between lesion types. Models with binary outcome of malignancy were studied using generalized estimating equations (GEE), (GEE), and using kinetic parameters and ADC values as malignancy predictors. Model accuracy was assessed using the corrected maximum quasi-likelihood under the independence confidence criterion (QICC). Predicted probability of malignancy was estimated for the best model. Results ADC values were significantly associated with kinetic parameters: medium and rapid uptake phase (p<0.001) and plateau and washout curve types (p=0.004). Comparison between lesion type showed significant differences for ADC (p=0.001), early phase (p<0.001), and curve type (p<0.001). The predicted probabilities of malignancy for the first ADC quartile (≤1.17×10−3 mm2/s) and persistent, plateau and washout curves, were 54.6%, 86.9%, and 97.8%, respectively, and for the third ADC quartile (≥1.51×10−3 mm2/s) were 3.2%, 15.5%, and 54.8%, respectively. The predicted probability of malignancy was less than 5% for 18.8% of the lesions and greater than 33% for 50.7% of the lesions (24/35 lesions, corresponding to a malignancy rate of 68.6%). Conclusion The best malignancy predictors were low ADCs and washout curves. ADC and kinetic parameters provide differentiated information on the microenvironment of the lesion, with joint models displaying improved predictive performance.

2015Journal article

Open access