ObjectiveTo explore the predictive value of MRI parameters and tumour characteristics before neoadjuvant chemotherapy (NAC) and to compare changes in tumour size and tumour apparent diffusion coefficient (ADC) during treatment, between patients who achieved pathological complete response (pCR) and those who did not.MethodsApproval by the Regional Ethics Committee and written informed consent were obtained. Thirty-one patients with invasive breast carcinoma scheduled for NAC were enrolled (mean age, 50.7; range, 37–72). Study design included MRI before treatment (Tp0), after four cycles of NAC (Tp1) and before surgery (Tp2). Data in pCR versus non-pCR groups were compared and cut-off values for pCR prediction were evaluated.ResultsBefore NAC, HER2 overexpression was the single significant predictor of pCR (p = 0.006). At Tp1 ADC, tumour size and changes in tumour size were all significantly different in the pCR and non-pCR groups. Using 1.42 × 10−3 mm2/s as the cut-off value for ADC, pCR was predicted with sensitivity and specificity of 88% and 80%, respectively. Using a cut-off value of 83% for tumour volume reduction, sensitivity and specificity for pCR were 91% and 80%.ConclusionADC, tumour size and tumour size reduction at Tp1 were strong independent predictors of pCR.

Pretreatment tumor ADC does not predict treatment response for patients with LABC undergoing NACT. Furthermore, ADC increase observed mid-way in the course of NACT does not correlate with tumor volume changes.

The purpose of this study was to assess the clinical value of a deep learning (DL) model for automatic detection and segmentation of brain metastases, in which a neural network is trained on four distinct MRI sequences using an input-level dropout layer, thus simulating the scenario of missing MRI sequences by training on the full set and all possible subsets of the input data. This retrospective, multicenter study, evaluated 165 patients with brain metastases. The proposed input-level dropout (ILD) model was trained on multisequence MRI from 100 patients and validated/tested on 10/55 patients, in which the test set was missing one of the four MRI sequences used for training. The segmentation results were compared with the performance of a state-of-the-art DeepLab V3 model. The MR sequences in the training set included pre-gadolinium and post-gadolinium (Gd) T1-weighted 3D fast spin echo, post-Gd T1-weighted inversion recovery (IR) prepped fast spoiled gradient echo, and 3D fluid attenuated inversion recovery (FLAIR), whereas the test set did not include the IR prepped image-series. The ground truth segmentations were established by experienced neuroradiologists. The results were evaluated using precision, recall, Intersection over union (IoU)-score and Dice score, and receiver operating characteristics (ROC) curve statistics, while the Wilcoxon rank sum test was used to compare the performance of the two neural networks. The area under the ROC curve (AUC), averaged across all test cases, was 0.989 ± 0.029 for the ILD-model and 0.989 ± 0.023 for the DeepLab V3 model (p = 0.62). The ILD-model showed a significantly higher Dice score (0.795 ± 0.104 vs. 0.774 ± 0.104, p = 0.017), and IoU-score (0.561 ± 0.225 vs. 0.492 ± 0.186, p < 0.001) compared to the DeepLab V3 model, and a significantly lower average false positive rate of 3.6/patient vs. 7.0/patient (p < 0.001) using a 10 mm3 lesion-size limit. The ILD-model, trained on all possible combinations of four MRI sequences, may facilitate accurate detection and segmentation of brain metastases on a multicenter basis, even when the test cohort is missing input MRI sequences.

ObjectivesTo explore how apparent diffusion coefficients (ADCs) in malignant breast lesions are affected by selection of b values in the monoexponential model and to compare ADCs with diffusion coefficients (Ds) obtained from the biexponential model.MethodsTwenty-four women (mean age 51.3 years) with locally advanced breast cancer were included in this study. Pre-treatment diffusion-weighted magnetic resonance imaging was performed using a 1.5-T system with b values of 0, 50, 100, 250 and 800 s/mm2. Thirteen different b value combinations were used to derive individual monoexponential ADC maps. All b values were used in the biexponential model.ResultsMedian ADC (including all b values) and D were 1.04 × 10-3 mm2/s (range 0.82–1.61 × 10-3 mm2/s) and 0.84 × 10-3 mm2/s (range 0.17–1.56 × 10-3 mm2/s), respectively. There was a strong positive correlation between ADCs and Ds. For clinically relevant b value combinations, maximum deviation between ADCs including and excluding low b values (<100 s/mm2) was 11.8 %.ConclusionSelection of b values strongly affects ADCs of malignant breast lesions. However, by excluding low b values, ADCs approach biexponential Ds, demonstrating that microperfusion influences the diffusion signal. Thus, care should be taken when ADC calculation includes low b values.Key Points• Diffusion-weighted sequences are increasingly used in breast magnetic resonance imaging• Diffusion-weighting (b) values strongly influence apparent diffusion coefficients of malignant lesions• Exclusion of low b values reduces the apparent diffusion coefficient• Flow-insensitive monoexponential apparent diffusion coefficients approach biexponential diffusion coefficients