Patient population
This prospective study was received approval from the Institutional Review Board (1802181-7). Between January and November 2018, 154 patients, who were diagnosed with breast cancer or suspected of having breast cancer underwent breast MRI in preparation to receive NAC. No biopsy or previous neoadjuvant treatment was performed before the baseline MRI. The diagnosis of breast lesions was confirmed by core needle biopsy and the diagnosis of a suspicious lymph node was confirmed by ultrasound-guided fine needle aspiration. The exclusion criteria included the following: patients with no obvious lesion detected on MRI (n = 1), with signal quality that was too poor to process DWI (n = 2), with pathology revealing lymphoma (n = 1), without pathology and loss to follow-up (n = 5). Among all the included patients, three were confirmed to have bilateral breast cancer. For multicentric or multifocal tumours, the tumours with the largest sizes according to MRI were analysed. Finally, 145 patients with 148 lesions were enrolled in this study.
Acquisition of DW images
All MRI was performed on a MAGNETOM Skyra 3 T MR system (Siemens Healthineers, Erlangen, Germany) using a dedicated 16-channel phased-array breast coil. The DWI in axial view was executed before contrast-agent injection using trace-weighted diffusion images (single-shot echo planar imaging) with spectral adiabatic inversion recovery for fat suppression were performed with the following parameters: b values (0, 10, 20, 50, 100, 150, 200, 400, 500, 800, 1000, 1500, 2000 s/mm2); repetition time/echo time, 5600/75 ms; flip angle, 90 degrees; field of view, 180 * 300 mm2; matrix, 96 * 200 mm2; slice thickness, 5.0 mm; 25 slices without gap; bandwidth, 1666 Hz; acquisition time, 6 min 21 s; generalized auto calibrating partially parallel acquisitions with an acceleration factor of 3; EPI factor, 96. The other sequences of breast MRI included a T1-weighted 2D gradient-echo and a fat-suppressed T2-weighted 2D fast spin-echo, as well as a fat-suppressed T1-weighted 3D fast spoiled gradient-echo sequence before and five times continuously after the contrast agent injection in the transverse plane, but these data were not considered for this study.
Postprocessing of IVIM and non-Gaussian diffusion data
DWI data were inline calculated by the scanner integrated Syngo software (Siemens Healthineers) according to the monoexponential, biexponential and kurtosis models. The monoexponential diffusion model was calculated by the following equation: Sb = S0 exp (−b* ADC) where ADC represents the apparent diffusion coefficient, and S0 and Sb are the signal intensity values in the voxels with b values of 0 and 1000 s/mm2, respectively. The bi-exponential model was expressed by the following equation: Sb/S0 = f exp(−b D*) + (1−f) exp(−b Dt), where Dt was the true diffusion, f was the perfusion fraction related to microcirculation and D* was the pseudo-diffusion coefficient which represents perfusion-related diffusion or incoherent microcirculation. The b-values used in IVIM are generally below 1000 s/mm2 (0, 10, 20, 50, 100, 150, 200, 400, 500 and 800 s/mm2). The non-Gaussian diffusion model was calculated by the following equation: Sb = S0 exp(−bMD + b2MD2 MKapp/6), where MD is the mean diffusivity and MK is the dimensionless metric mean kurtosis expressing the deviation from the Gaussian distribution. The b-values used in IVIM are generally high b values (0, 500, 1000, 1500 and 2000 s/mm2). Parametric maps, including ADC1000, Dt, f, D*, MK and MD maps, derived from the three diffusion models were generated with least squares fitting of all b-value data on a voxel-by-voxel basis with software.
Volumetric-tumour histogram-based analysis in IVIM and non-Gaussian diffusion parameters
Histogram analysis was performed with the prototype MR Multiparameter Analysis software (Siemens Healthineers). The analysis of DWI-derived IVIM and non-Gaussian diffusion parameter maps were executed separately. Regions of interest (ROIs) were placed manually on the DW images with a b value of 1000 s/mm2, but DCE images to assist in locating the lesions and verifying the lesion boundaries. ROIs were placed on all slices that contained the whole tumour and the largest lesion (in the case of multicentric or multifocal tumours), and care was taken to avoid regions influenced by partial volume effect (Fig. 1). Two radiologists (C.Y. and Y.Q.C. with 6 and 2 years of experience in breast MRI, respectively) were blinded to the pathological and biochemical findings of each patient, and reviewed the MR images and draw the ROIs independently. When discrepancy of ROIs arose especially for non-mass enhancement lesion, two of them together made a consensus of lesion and redraw the ROIs later. The mean ROI of lesion for radiologist 1 was 37.72 ± 77.10, and the mean ROI of lesion for radiologist 2 was 45.24 ± 84.64. Spearman correlation showed ROI of tumour on DW image had good agreement with two radiologists (r = 0.835, p < 0.001). Finally, the data from the two radiologists’ average measurements were analysed.
Histogram analysis for the whole tumour on the parametric maps was performed and the parameters were extracted, including percentiles (5th, 50th and 95th of the ADC value), skewness (a measure of asymmetry of the probability distribution), kurtosis (a measure of the shape of the probability distribution), contrast (a measure of the signal difference) and entropy (a measure of texture irregularity).
Histopathological analysis
All pathological results were defined according to the World Health Organization classification of breast tumor [17]. According to IHC-determined steroid HRs with estrogen receptor (ER), progesterone receptor (PR), and HER2 status, as well as tumour proliferation measured by Ki 67, breast cancer is considered to consist of four molecular types: (1) luminal A-like subtype (ER or PR positive, or both, HER2 negative, low proliferation); (2) luminal B-like subtype (ER or PR positive, or both, HER2 negative, high proliferation); (3) HER2 subtype, non-luminal (HER2 positive and ER and PR negative) or luminal (HER2 positive and ER or PR positive, or both); (4) basal-like subtype (HER2 negative and ER and PR negative; i.e. triple-negative breast cancer) [18].
Statistical analysis
All data were analysed using SPSS 20.0 (Chicago, IL). Values of p < 0.05 were considered statistically significant. Categorical data were compared with the Pearson Chi squared test. IVIM and non-Gaussian diffusion parameters in the status of molecular prognostic biomarkers were compared by Student’s t test when normally distributed or by the Mann–Whitney U test when not normally distributed. Receiver operating characteristic (ROC) curve analysis was used to evaluate the effectiveness of IVIM and non-Gaussian diffusion parameters for differentiating HER2 positive breast cancer. The area under the curve (AUC), sensitivity, and specificity at the best cut-off point were reported. The Spearman correlation coefficient was calculated to analyse the correlations between of clinical TNM stage, and Ki 67 status with IVIM and non-Gaussian diffusion parameters.