This prospective study was approved by our Institutional Review Board and compliant with the Health Insurance Portability and Accountability Act (HIPAA). Patients with stage II/III breast cancer who were scheduled to undergo NAC were enrolled in this study (NCT01931709, registration date August 28, 2013) from May 2012 to July 2015. Patients with distant disease beyond regional lymph nodes were excluded. Standard clinical pathological assessment included immunohistochemistry analysis of diagnostic core-needle biopsy specimens for determination of breast cancer subtype based on estrogen receptor (ER), progesterone receptor (PR), and HER2 status (by immunohistochemistry and/or fluorescence in situ hybridization), and Ki-67 proliferation index [16]. Patients were imaged with MRI and 18F-FDG PET prior to initiation of treatment, during the NAC (2–12 weeks), and after the completion of NAC.

Patients were imaged on a 3 T Philips Achieva Tx MRI scanner (Philips Healthcare, Best, The Netherlands) equipped with a dedicated 16-channel bilateral breast coil (Mammo-Trak, Philips Healthcare, Best, The Netherlands). Multiparametric breast MRI exams were obtained in the axial orientation and included DW-, and T1-weighted DCE-MRI sequences. DW-MRI was acquired with a single-shot echo-planar imaging sequence with fat suppression, with b values of 0, 100, and 800 s/mm2. DCE-MRI was acquired with a fat suppressed 3D fast gradient echo (eTHRIVE) sequence, with T1-weighted images were collected before and after administration of gadolinium-based contrast agent (ProHance, bracco Diagnostics, Milan, Italy) at 0.1 mmol/kg body weight. Post-contrast sequences were acquired with k-space centered at 2, 5, and 8 min after contrast injection. Additional details regarding image acquisition are provided in the Additional file 1.

18F-FDG radiotracer was purchased commercially (Cardinal Health, Seattle, WA) or produced in house. PET imaging was performed on an GE Discovery STE PET/CT scanner (GE Medical Systems, Waukesha, WI) as previously described for dynamic imaging [12, 17] with a low dose CT for attenuation correction and positioning. Dynamic imaging was performed over the chest and breast for 60 min after the start of FDG infusion [7–11 mCi (259–407 MBq)] and was followed by a clinical protocol of 5 fields-of-view static imaging. Summed standardized uptake value (SUV) images from 30 to 60 min post injection were constructed from the dynamic data. Additional details regarding image acquisition are provided in the Additional file 1.

As previously described [11, 18], DCE-MRI data was analyzed using custom software written in Java and ImageJ (NIH, public domain, Bethesda, MD), providing voxel-wise measures of percent peak enhancement (PE) and signal enhancement ratio (SER). PE was calculated as (S1 − S0)/S0, where S0 and S1 are the pre-contrast and early (2 min) post-contrast signal intensities, respectively. SER was calculated at each voxel, as (S1 − S0)/ (S2 − S0), where S2 is the delayed phase post-contrast image (8 min).

Tumors were segmented in 3D based on percent enhancement of 50% or greater at 2 min post-contrast. A hotspot analysis was performed to identify peak PE and peak SER for each lesion, where a peak was defined as the highest mean value for a 3 × 3 voxel tumor subregion. Additionally, the functional tumor volume (FTV, cc), i.e., the total volume of voxels with PE ≥ 50%, and washout volume (cc), i.e., the total volume of voxels with SER ≥ 1.1, were calculated for each tumor.

DW-MRI data was fit to the conventional monoexponential decay model to calculate the apparent diffusion coefficient (ADC) at each voxel using custom MATLAB-based software (Mathworks, Natick MA). Post-contrast T1-weighted images were used to localize lesions, prior to segmentation of tumor regions-of-interest (ROIs) on the b = 800 mm2/s image. Lesion ROIs were applied to ADC maps and used to calculate mean ADC for each lesion at each time point.

To calculate SUVmax, using the 30–60 min summed images constructed from the dynamic data, volumes-of-interest (VOIs) of approximately 1 cc were drawn on lesions identified on the pre-therapy PET scan encompassing the pixels with the most uptake. Using the pixel with the most uptake, SUVmax was calculated as the maximum tissue activity divided by injected dose/body weight.

Dynamic imaging and kinetic modeling were done as previously described [12, 17]. Briefly, the VOIs drawn on the 30–60 min summed images were applied to the dynamic image set. An approximately 1 cc VOI was also drawn over the left ventricle to create the blood input function. A two-tissue compartment model was utilized to calculate the kinetic parameters using PMOD version 3.6 (Zurich, Switzerland). Metabolic flux (Ki), was estimated from parameters derived by fitting the input function and the blood-activity curve to the tissue time-activity curve data, and calculated as (K1 * k3)/(k2 + k3), where K1 represents the transfer of blood into tissue, k2 is the transport back to blood, and k3 represents metabolic trapping of the tracer. Finally, the metabolic rate of FDG (MRFDG) was calculated as Ki* [Glucose].

Pathological response was determined after NAC through histopathological evaluation of the surgical breast specimen by a breast pathologist. Residual cancer burden (RCB) was assessed, with patients categorized as RCB class 0, I, II, or III using methods described by Symmans et al. [19]. Pathological complete response (RCB 0) was defined as no residual invasive disease in the breast or lymph nodes. Following STEEP criteria, RFS was defined as the time between initiation of NAC and disease recurrence or death, whichever came first [20].

Metabolism/perfusion ratios were calculated as the ratio of MRFDG or SUVmax to peak SER or peak PE (i.e., MRFDG/peak SER, MRFDG/peak PE, SUVmax/peak SER, SUVmax/peak PE). Percent change from baseline were calculated for MRI and PET parameters at mid-treatment and post-treatment time points. Spearman rank-order correlation coefficients were calculated between baseline MRI and PET measures. The Wilcoxon rank-sum test was used to compare the percentage change in imaging measures for patients with RCB 0/I versus RCB II/III after NAC. Patients who developed metastases during NAC and did not proceed to surgery were grouped with RCB II/III patients. Univariate Cox proportional hazards regression was used to examine the association of percentage change in each imaging parameter with RFS, with the Wald test used to evaluate each parameter’s estimated hazard ratio. To correct for multiple comparisons, the Benjamini–Hochberg procedure was used to correct p values from Wilcoxon rank-sum tests and Cox proportional hazards analyses. Kaplan–Meier curves for RFS were computed for patients dichotomized into two groups based on the third quartile (i.e., 75th percentile) of the percentage change for a given image parameter. Log-rank tests were used to compare survival curves. All statistical analyses were performed in R (R Foundation for Statistical Computing, Vienna, Austria). A p value < 0.05 was considered significant.