Figure 4. Comparison between DW images acquired using a series of b-encoding gradients (right-hand panels), and DCE subtraction images on the same patient. Diffusion encoding b-values are indicated on top of each column for SPEN DW images. In the two left columns T2w TSE and T1w DCE subtraction images are also presented. Shown are exams for four IDC (rows 1, 4, 5 and 6) and two ILC (rows 2 and 3) patients. These images suggests the possibility to identify and delineate lesions solely on the basis of strongly b-weighted images –without a need for contrast and for DCE subtractions.

Figure 2. Representative ADC and kurtosis maps obtained from SPEN and RESOLVE data for four patients, displaying three IDC breast lesions and one ILC lesion (bottom left panel). Shown for each patient are single-breast anatomical T2w TSE and T1w DCE subtraction images, with the latter highlighting as bright regions the cancerous masses. Shown as well are the kurtosis and the ADC maps derived by the SPEN and RESOLVE exams (ADC maps were calculated solely on the basis of images acquired with 0 and 850 s/mm² nominal b-values).

Figure 1. Example of the direction averaged correction map Cave and impact on tumor ADC values. A 50 year old woman with a 7 mm suspicious BIRADS 4 lesion, which was found to be malignant on biopsy. (a,b,c) show the uncorrected ADC map, GNL correction map and the resulting ADC map after applying GNL correction. Insets on images (a) and (c) show lesion area. The lesion ADC values were lower after GNL correction. (Hotspot ADCs Uncorrected = 1.19 x10^-3 mm²/s Corrected = 0.99 x10^-3 mm²/s)

Figure 2. Bland-Altman plot for uncorrected and corrected hotspot ADCs by MRI gradient system

Figure 1: I-SPY 2 treatment de-escalation strategy

The treatment de-escalation decision will be made at T2 based on combined MRI and biopsy results. Qualified subjects will be given the option to skip anthracycline-cyclophosphamide (AC) treatment and proceed directly to surgery.

Figure 3: Predictive performance of pCR in the optimal and non-optimal segmentation groups (Main analysis: Reader 1)

AUC values for the prediction of pCR were higher for the optimal segmentation group versus non-optimal group for ∆FTV1, ∆FTV2 and the multivariable model (0.68 vs. 0.66, 0.70 vs. 0.62, and 0.84 vs. 0.64, respectively), with the difference reaching statistical significance for the multivariable model.

-ROC analysis for diagnosing pCR based on DWI score/Kinetic score-

Kinetic score showed slightly higher AUC while 95% confidence interval overlapped with that of DWI score. Both kinetic score and DWI score demonstrated excellent diagnostic performance among triple negative subtype compared to other subtype with AUC of 0.88-0.95. For luminal subtype, DWI score tended to perform better than kinetic score.

-Image evaluation-

• DWI score

DWI of the target lesion was evaluated and categorized as 3-point scale.

0: no abnormal signal, 1: small focus of high - intermediate signal intensity, 2: obvious high signal intensity.

• Kinetic score

Kinetic patterns of the lesions on DCE-MRI was scored as

0 : no enhancement, 1 : persistent, 2 : plateau and 3 : washout.

For both scores Low score indicates pCR.

Figure 3. Example cases

Figure 1: A 41-year-old patient with a DCIS. (a) F1 Pre-contrast image. (b) F2 post-contrast image. (c-i): The zoom-in smallest bounding box containing the tumor. (c) F1 pre-contrast, (d) F2 post-contrast, (e) F3 post-contrast, (f) The last F6 post-contrast image, showing a comparable enhancement as in F3. (g) The wash-in signal enhancement map F2-F1, (h) The maximum F3-F1 signal enhancement map, (i) The wash-out F6-F3 map. A small portion of the tumor shows the wash-out pattern. (J) The DCE time course shows a plateau pattern, after reaching the maximum in F3.

Figure 2: A 63-year-old patient with an invasive ductal cancer (IDC). (a) F1 Pre-contrast image. (b) F2 post-contrast image. (c) F1 pre-contrast, (d) F2 post-contrast, (e) F3 post-contrast, (f) The last F6 post-contrast image, showing wash-out DCE pattern with decreased intensity after reaching maximum in F3. (g) The wash-in signal enhancement map F2-F1, (h) The maximum F3-F1 signal enhancement map, (i) The wash-out F6-F3 map. (j) The DCE time course shows a typical wash-out pattern, reaching maximum in F3, followed by decreased intensity from F4 to F6.

Figure 1 Flow chart. A tumor region was contoured manually on T1W_{post 90s}­, to which other images were aligned. First order and texture features were extracted from T1W, DCE kinetic maps, T2W, and ADC maps, and used to build radiomics models, together with shape features. The dataset was split into training and test cohort by scanning date. After feature selection, radiomics models were constructed using SVM or LR, before evaluated with ROC analysis, DCA

Figure 2 Model evaluation. (a) ROC in internel test cohort. (b) Decision curve analysis for each model in the testing dataset. The decision curve showed that when the threshold probability is in the range 0.85-0.95, the application of the final model adds more benefit than treating all or none of the patients and other models. (c) Calibration curve of the union model prediction in the internal and external test cohort. (d) Weights of selected features in the final model

Figure 1. 3D Residual Attention U-Net architecture for brain lesion data. The network consists of 6 encoding layers (purple) and 6 decoding layers (orange). Spatial dimension decreases by 2 and channel dimension increases by 2 as the data propagates through the encoding layers while the reverse happens along the decoding layers. A full scan is then returned as the model output, as the prediction of the entire scan.

Figure 2. 2D Residual Attention U-Net architecture for breast lesion data. The network consists of 5 encoding layers (purple) and 5 decoding layers (orange). Spatial dimension decreases by 2 and channel dimension increases by 2 as the data propagates through the encoding layers while the reverse happens along the decoding layers. A single slice of the predicted scan is then returned as the model output.

Fig. 3: Multimodal images of a 33-y.o. breast cancer patient diagnosed of high-grade HER2+ invasive ductal carcinoma measured 4.5×2.6×2.6 cm. (a) T1-weighted fat saturated (FS) post-contrast MRI. (b) DOT image of total hemoglobin concentration (HbT) overlaid with simultaneously acquired T1 non-FS MRI. (c) Image of shear modulus measured by MRE. Red line – tumor marking. White dotted line – MRE actuator contact.

Fig. 2: Multimodal MRE/DOT imaging results of a dual-contrast tissue phantom. (a) T1 image showing three 20-mm diameter inclusions marked in circles. (b) MRE phase image obtained using 100Hz vibration showing wave propagation within the entire 12-cm diameter phantom with longer wavelengths inside inclusions. (c) Reconstructed shear modulus map and (d) absorption coefficient map shows clear contrasts in expected inclusion locations. Inclusion 3 was out of the optical coverage marked by the vertical dotted line in subplot (d), resulting in failure to recover its optical contrast.

Figure 1: Example T1, T2, and PD maps processed from the MAGIC sequence using SyMRI

Figure 2: The variable importance features within the T2 radiomic model

Figure 1: A 50-year-old patient with a malignant cancer showing smooth boundary. (a) F1 Precontrast image. (b) The F2 postcontrast image. (c) The F1 precontrast image. (d) The F2 postcontrast image. (e) The F3 postcontrast image. (f) The last F6 postcontrast image,showing persistent enhancement with increased intensity over time. (g) The washin signal enhancement map F2-F1. (h) The F3-F1 signal enhancement map. (i) The washout F6-F3 map. (j) The corresponding CC view mammography, the lesion mass was outlined by green line.

Figure 2: A 58-year-old patient with a malignant cancer showing smooth boundary. (a) F1 Precontrast image. (b) The F2 postcontrast image. (c) The F1 precontrast image. (d) The F2 postcontrast image. (e) The F3 postcontrast image. (f) The last F6 postcontrast image, showing persistent enhancement with increased intensity over time. (g) The washin signal enhancement map F2-F1. (h) The F3-F1 signal enhancement map. (i) The washout F6-F3 map. (j) The corresponding CC view mammography, the lesion mass was outlined by green line.

Figure 1. The diagram of the proposed two-step deep learning model.

Figure 3. ROC curves of 3 models on testing data

Table 1: The accuracy of breast cancer grading using three classifiers according to different feature sets

Figure 1: Representative images of Grade Ⅰ-Ⅲ breast cancer. The arrows point to the breast cancer lesions.

Figure 1: The flowchart of the experimental design. The tumor is segmented by Fuzzy-C-means clustering algorithm on F2 post-contrast image, and then the tumor ROI is mapped to 3 generated DCE parametric maps. On each map, 32 first-order and 75 texture parameters are extracted using the PyRadiomics. For each case, a total of 268 radiomics features with ICC ³ 0.8 are used to build models using five machine learning algorithms to differentiate three different molecular subtypes.

Figure 2: A 48-year-old patient with an invasive ductal cancer (TN). (a) F1 Pre-contrast image. (b) F2 post-contrast image. (c-i): The zoom-in smallest bounding box containing the tumor. (c) F1 pre-contrast, (d) F2 post-contrast, (e) F3 post-contrast, (f) The last F6 post-contrast image. (g) The wash-in signal enhancement map F2-F1, (h) The maximum F3-F1 signal enhancement map, (i) The wash-out F6-F3 map. (J) The DCE time course shows a typical wash-out pattern.

Figure-1: The flow chart of the proposed framework.

Table-1: List of different feature vectors and their description.

Fig. 3. Forest plot of single studies for the pooled diagnostic odds ratio and 95% CI

Figure 1. Illustration of the deep learning ensemble developed for pathologic complete response (pCR) prediction for the triple-negative breast cancer cohort. The ensemble took the pre-treatment DCE and DWI as input. Two convolutional neural networks extracted features from the DCE and DWI, respectively. The sequences of features were then input to the two recursive neural networks, respectively. Outputs of the recursive neural networks were concatenated and used for pCR or non-pCR prediction.

Figure 3. Receiver operating characteristic curve (blue) of the prediction using the deep learning ensemble. Using the pre-treatment DCE and DWI, the ensemble achieved the best accuracy of 69%, with the sensitivity of 75% for pCR patients and specificity of 63% for non-pCR patients. The area under the curve (AUC) was 0.68.

Figure 5: True negative (TN) case example from a 44-year-old patient with a confirmed benign adenosis in the left breast. Extensive parenchymal enhancements are seen in both breasts. (a) Pre-contrast image acquired using fat-sat sequence; (b) Post-contrast image; (c) Tumor detection results searched by the Mask R-CNN algorithm. Two boxes are generated to identify two suspicious lesions, one in each breast. After evaluation by ResNet50 network, the left lesion has a malignant probability of 0.44, thus correctly diagnosed as benign. The parenchymal enhancements from normal tissues in the right breast has a very low malignant probability of 0.12. These results illustrate the potential of Mask R-CNN detection followed by ResNet50 classification for automatic detection and further characterization of identified lesions to develop a fully-automatic computer-aided diagnosis system for breast MRI.

Figure 1: Flow diagram of the training and testing courses using Mask R-CNN for detection (shown in purple) and ResNet50 for classification (shown in blue).

Fig. 1 Methods of FGT ratio calculation. (a) Example slices of pre-contrast image (top), whole breast (middle) and FGT segmentations (bottom) for one typical exam labelled ‘extreme fibroglandular tissue’ according to the radiology report. (b) Ratio of FGT (orange) to whole breast (blue) segmentations is higher for center slices (ρ=49%, inset) compared to all slices (ρ=36%). (c) Resulting FGT segmentation following the Maximal Intensity Projection method skews densities even more toward higher values (ρ=70%).

Fig. 2. Distribution of FGT amounts per radiology-defined class for different density calculation methods. (a) Simple density, where all slices were considered for the FGT amount calculation, (b) threshold-based density, where only slices with a significant amount of FGT are considered, and (c) maximal intensity projection, where segmentations across all slices were projected onto a 2-dimensional array prior to the density calculation.

Figure 1. Representative images of breast distortion phantom polycarbonate grid collected with full-FOV EPI with (left) and without (middle) parallel imaging (PI) and reduced-FOV EPI without PI (right). Images of positive (A-C) and negative (D-F) PE direction before RPG distortion correction. (G-I) Positive PE direction images after RPG correction. (J-L) Overlay of each circle’s center location from DWI data before (magenta) and after (green) RPG on anatomical reference image.

Figure 2. Magnitudes of distortion artifacts before and after RPG correction in (A,B) the phase-encoding, (C,D) frequency-encoding directions for full-FOV EPI with (red) and without (green) parallel imaging (PI) and reduced-FOV EPI without PI (blue). Horizontal black bars indicate p<0.05 significance.

Table 2: ADC measures for pre and post-contrast DWI

Figure 1: Bland-altman plot for all lesions, comparing pre and post-contrast ADC

Figure 7. The area under the ROC curve of the mean ADC, IVIM-D, IVIM-DP, DKI-K and DKI-D values are 0.915, 0.909, 0.574, 0.768 and 0.895, which indicates that the ADC value is the best single quantitative parameters to distinguish benign and malignant breast lesions. The area under the ROC curve of combined ADC and DKI-K value is 0.923.

Table2:Diagnostic Effectiveness for Benign and Malignant Lesions of Quantitative Parameters

Figure 2:

the representative case of the typical malignant tumor (left). The tumor shows moderate signal intensity at low b-value, an increased signal at high b-value, and decreased ADC, suggesting malignancy. Invasive ductal carcinoma.

the representative case of the typical benign tumor (right). The tumor shows marked signal intensity at low b-value, decreased signal at high b-value, and slightly low ADC, suggesting a benign breast tumor. Fibroadenoma.

Table 2: Diagnostic performance for DWI and DCE-MRI

Table5 and Figure show ROC curve of the parameters in Table 5 and parameters in forecasting performance of model

Table 4.Comparison of the change rate of parameters in histopathological response groups during and before NAT

Figure 1. (A) Diffusion breast phantom schematic. Percentages indicate % w/w PVP in water. (B) Schematic with tubes and fibroglandular ROI locations. (C) Axial view of ADC map. Yellow line shows the slice at which ROIs of water (black) and fibroglandular tissue (pink) were drawn in the (D) coronal view. Green line shows the slice at which (C) was taken. (E) Median ADC, shown as dots for each tube. Red lines indicate expected values.^6,8

Figure 2. Coefficient of variation across scanners of (A) absolute ADC, and (B) ADC relative to water for each tube.

Figure 2: DTI λ1 parametric map (A) and maximum intensity projection (MIP) of the DCE images (B) in a lactating 33 years old BRCA-1 carrier woman. The λ1 parametric map depicted in the left breast a non-palpable invasive ductal carcinoma lesion (low λ1 values) that was detected by screening DTI but was not demonstrated by DCE because of the masking effect of marked background parenchymal enhancement (BIRADS 1 DCE and BIRADS 5 DTI).

Figure 1: DTI λ1 parametric map (A) and maximum intensity projection (MIP) of the DCE images (B) in a lactating 33 years old breast cancer patient. The λ1 parametric map and the DCE images both show in the left breast a triple negative invasive ductal carcinoma mass lesion (low λ1 values ). The cancer lesion was diagnosed by both DCE (assigned BIRADS 4) and by DTI (assigned BIRADS 5).

Figure 1: Axial T2w breast MR images reconstructed with deep learning scored significantly higher in SNR and image sharpness than those without deep learning. (a,d) A patient with substantial fibroglandular tissue; (b,e) a patient with multiple simple and complicate cysts; and (c,f) a lactating patient.

Figure 3: T2w images acquired at 0.714 x 0.714 mm² resolution (a) appear noisier than those acquired at the standard 1.1 x 1.1 mm² resolution (c). However, application of DL (b) increases SNR to achieve SNR more similar to the lower spatial resolution protocol.

Table1: Diffusion Parameters of molecular subtype

Table2: Diffusion Parameters of molecular prognostic factors

Figure. 1 Synthetic MR, DCE-MRI and DWI images of a 42-year-old woman with Invasive cancer in the left breast. the synthetic images obtained 11 minutes after contrast agent injection. A T2WI (Synthetic)，the position of the ROI is outlined with a red circle; B T1 map (Synthetic); C T2 map (Synthetic); D proton density map (Synthetic); E DCE-MRI (2 min 31s after contrast injection); F ADC map (DWI).

Figure. 2 Synthetic MR, DCE-MRI and DWI images of a 39-year-old woman with adenosis in the left breast. the synthetic images obtained 11 minutes after contrast agent injection. A T2WI (Synthetic), the position of the ROI is outlined with a red circle; B T1map (Synthetic); C T2map (Synthetic); D proton density map (Synthetic); E DCE-MRI (2 min 31s after contrast injection); F ADC map (DWI).

Figure 3. The DW image findings are different among 3 readers. Fibroadenoma.

Table 1. Agreement in DW-based BI-RADS categories. Agreement in DW-based BI-RADS categories tended to be higher in b1500 compared to b800 DW images. Mean and confidence intervals for kappa values are shown.

Fig 1. The ROC of T2_10th, T2_mean, T2_median and T2_90th values distinguishing TNBC from luminal A subtypes

Fig 2. The ROC of T2^_10th, T2_mean, T2_median and T2_90th values distinguishing TNBC from luminal B subtypes

Figure 2. Comparison of QTM method and Kety’s method on a malignant lesion. This is a 73 years old patient with biopsy proven malignant lesion. a) post-Gd T1 weighted image, b) QTM |u| map, c) K^trans and d) V_e map using internal mammary (IM) AIF.

Figure 3. Differentiating malignant breast lesions from benign breast lesions. a) QTM |u| (p<0.001), b) K^trans (p=0.007), c) V_e (p=0.006) and d) A (p=0.001) demonstrating significance difference between malignant and benign lesions. There were no other parameters demonstrating significant difference between malignant and benign lesions.

Figure 4: Individual time frames at peak contrast enhancement (time = 240 s) reconstructed using SENSE (A), CS-TV (B), MOCCO with model order K=3 (C) with temporal resolution of 15 s. High image quality is observed for both regularized reconstructions (B, C). Note the late-phase image (D) was acquired with a navigator-gated Cartesian acquisition, which resulted in a longer acquisition time (8:43 min:s).

Figure 5: Supine image (A) from a patient volunteer reconstructed using SENSE (green, X), CS-TV (blue, circles), and MOCCO with model order K=3 (red, stars) with 15 s temporal resolution. PSC (%) are plotted from ROIs placed in the lesion, fibroglandular tissue, and aorta (A, red outlines). Rapid wash-in and wash-out contrast kinetics are observed in the aorta D). The lesion B) showed relatively rapid contrast update, while the fibroglandular tissue C) showed slower contrast uptake.

Figure 2: Examples of the SFA maps in contra- and ipsilateral breast and small ROI around the tumor. Only the voxels with >90% fat fraction were included in the maps.

Figure 3: SFA comparison between contralateral, ipsilateral breast, and a small ROI around the lesions for patients with (a) benign and (b) malignant lesions. A significantly higher (p= 0.007) SFA was observed around the malignant lesions.

Fig1 ROC curve of IVIM parameters D, D* and F for the diagnosis of neoadjuvant chemotherapy in breast cancer

Fig2 A-H A 64-year-old woman has a large tumor on her right breast. The tumor is isointensity on T1WI, slight hyperintensity on T2WI, and heterogeneous enhancement. E, F, G, and H: The value of ADC, D, D*, and f is 1.04×10-3 mm2/s, 0.905×10-3 mm2/s, 23.1×10-3 mm2/s, 9.94%.

Figure 1. Study design.

A two-group cross-sectional study in a flow chart. Eight specimens were excluded due to small tumour size and mixed phenotype. Regions of interests (ROI) were drawn on chemical shift-encoded images (CSEI) to define the adipose tissue boundary. Fat, water and the number of double bonds in triglyceride molecules were estimated from the CSEI data, from which saturated, monounsaturated and polyunsaturated fatty acids (SFA, MUFA and PUFA) were derived. Statistical comparison was conducted on lipid components between Scores 2 and 3 tubule formation.

Figure 2. Group differences in peri-tumoural monounsaturated fatty acids (MUFA) in breast adipose tissue.

The difference in MUFA (a) mean, (b) skewness, (c) entropy and (d) kurtosis between breast cancer with Scores 2 and 3 tubule formation are shown in dot plots. Each dot represents the spatial distribution around the breast tumour, and the dots are organised in two columns corresponding to the two groups. The error bar indicates the mean and standard deviation. The t-tests were performed and p value is shown for each plot. Statistically significant p values (< 0.05) are marked by ‘*’.

Table 2 ROC curve analysis of quantitative and semi quantitative parameters of benign and malignant breast lesions groups

Figure.2 the ROC curve of Ktrans, CER, IAUGC, Ve, MaxSlope distinguishing between benign and malignant breast lesions groups

FIGURE 1. ROC curves of different methods (DCE-MRI, IVIM, and DCE-MRI + IVIM) for discriminating malignant and benign breast lesions.

TABLE 1. The correction analysis of DCE-MRI, IVIM and the prognosis of breast cancer indexes.

This chart illustrates the flow of data into the network.

1) The first three phases of each DCE Exam are gathered and stacked in the 4th dimension.

2) Slices containing the biopsy site are then isolated and annotated.

3) Before introduction to the model, the data is exposed to different scales of patch sampling, Hue shifts, and standard affine transforms.

4) The network is trained using 5 fold cross validation with leave one out testing

Performance Statistics for various sampling techniques

Figure 1. Pipeline for registration-based breast segmentation algorithm

Figure 3. Registration-based breast segmentation method better identifies the lower boundary of the breast region. The resulting segmentations (red) are overlaid to the breast images with the problematic area circled in green.

Fig. 5: Comparison of breast DWI of a healthy volunteer. Upper row: standard DWI based on EPI, lower row: DWI based on DESS (dwDESS). The b-value for dwDESS is calculated using the proposed method and leads to same image contrast as standard DWI.

Fig. 2: Overview of the diffusion signal model for a double-echo steady-state sequence in the bipolar case (details see text)

Figure 4. Visualization of error maps of (A) K^trans = 0.133, 0.399, 0.666, 1.602 min^-1 and V_e = 0.3 obtained by measuring the % differences between the fitted parameters and the true values (% error) for the lesion region depicted in (B). Red and blue represent the level of overestimation and underestimation, respectively.

Figure 2: Simulated contrast agent concentration uptake curves (displayed for a subset of time from 150 s to 400 s). Mean concentration for eight lesions with varying pharmacokinetics reconstructed using MOCCO (A-H, red squares). Standard deviations are shown with banded areas. The input time curves (“truth”) used to generate the source data are plotted in black for all frames.

Fig 1: Example ADC maps from standard DWI scan, showing full acquisition (top row) and retrospectively shortened scans (middle and bottom row). The columns show the original scan (left), with low-rank denoising (middle), and the difference (right). Note the low-ADC tumor indicated by the red arrow.

Fig 2: Patch-based analysis showing the impact of NORDIC on ADC maps as a function of SNR. For this example, all 40 ADC slices from a single subject were divided into 12x12 pixel patches, and the mean absolute difference between original and denoised patches was plotted as a function of SNR. The black line is a filtered moving average. In the full acquisition (A) the greatest impact of denoising is for patches with SNRdB<10, with the effect decreasing with increasing SNRdB. In the abbreviated acquisitions (B, C) the overall SNR is lower and more patches are affected by the denoising operation.

Data summarizing the spleen weights in (A) SCID mouse groups with non-tumor-bearing control (n=6), SUM-149-EV (n=9), SUM-149-COX-2 (n=12) mice and (B) BALB/c mouse groups with non-tumor-bearing control (n=5), 4T1-wt (n=3), 4T1-EV (n=5) and 4T1-COX2shRNA (n=5) mice. Values represent Mean ± SEM.

Intensity quantification normalized to SCID mouse spleen weight from non-tumor-bearing control (n=6), SUM-149-EV (n=9), SUM-149-COX-2 (n=12) groups illustrated respectively for (A) glutamate, (B) glutamine, (C) glutathione, (D) lactate and (E) aspartate. Values represent Mean ± SEM.

Figure 1. Study design.

A two-group cross sectional study in a flow chart. Freshly excised breast tumours from wide local excision or mastectomy were immediately scanned on a clinical 3 T MRI scanner to derive lipid composition using double quantum filtered–correlation spectroscopy (DQF-COSY). Immunohistochemical examinations were conducted to assess lymphovascular invasion (LVI), Ki-67 expression and Nottingham Prognostic Index (NPI). In total, 30 patients with invasive ductal carcinoma (IDC), 13 with LVI negative and 17 with LVI positive, participated in the study.

Figure 2. Group difference in lipid composition.

The difference in (a) monounsaturated fatty acids (MUFA), (b) triglycerides (TRG), (c) saturated FA (SFA) and (d) polyunsaturated FA (PUFA), are shown in dot plots. Each dot represents the measurement from a patient, and the dots are organised in two columns corresponding to the lymphovascular invasion (LVI) status. The error bar indicates the median and interquartile range. The Mann Whitney U tests were performed between the groups and p value is shown for each plot. Statistically significant p values (< 0.05) are marked by ‘*’.

Figure 1: (A,B) T₁ images from a signal patient acquired throughout neoadjuvant chemotherapy treatment. (C,E) Images from follow-up visits were registered to baseline using FLIRE and (D,F) a well established registration program, DRAMMS. (E,F) Registered images are overlaid with baseline to evaluate differences. White indicates image similarities, blue is unique to baseline, and red is unique to the registered images. Both registration methods adequately correct for displacements in breast tissue.

Figure 3: To compare registration accuracy, similarity measures for each patient were averaged across time points. In the box plot, the red line indicates the median value across all patients, the box indicates the range of values between the 25^th and 75^th percentile, the whiskers indicate minimum and maximum values within 1.5 times the interquartile range, and the red plus signs indicate values beyond this range.

Figure4. A representative case of f_IVIM map at long diffusion time (A; in vivo, B; post-mortem and C: CD31 staining). f_IVIM values dropped post-mortem as expected, due to the lack of perfusion, but did not reach 0. The lower f_IVIM area was well correlated with the CD31-negative non-brownish area (arrow).

Figure1. Box-whisker plots and line graphs of f_IVIM values with different diffusion times in vivo and post-mortem.

CEST-mDixon result. Left: MTR_asym maps for 1ppm, 2ppm and 3.5ppm for 3 slices. Right: Z-spectrum from the ROI placed over the suspicious lesion.

Left to right: ROI averaged MTRasym­(1ppm), MTR_asym­(2ppm) and MTR_asym­(3.5ppm) compared to the Ki-67 histopathological index. A moderate positive correlation is observed for hydroxyls (1ppm) and amines (2 ppm).

Figure 1: A slice through the center of two invasive ductal carcinomas is shown in Fig. 1, for T2W-VISTA (a) and HiSS MRI water peak height (b) images. (c) and (d) show conductivity maps derived from (a) and (b), respectively, using EPT.