Pretreatment dual-energy CT versus diffusion-weighted imaging for predicting pathologic complete response to neoadjuvant chemotherapy in breast cancer.

Background
Neoadjuvant chemotherapy (NAC) has been established as the standard treatment for locally advanced breast cancer [1], offering the advantages of drug sensitivity detection, tumor downstaging, and increased potential for breast-conserving surgery [2]. Achieving pathologic complete response (pCR) after NAC is associated with improved overall and recurrence-free survival [3]. However, response to NAC varies across breast cancer molecular subtypes, with pCR rates ranging from 0.3% to 50.3% [4]. Current evaluation of pCR relies on invasive pathologic procedures of postsurgical specimens, which increase the risk of delayed intervention in non-responders. Therefore, pre-NAC prediction of pCR may facilitate an improved clinical outcome by enabling timely treatment modification and supporting personalized patient management.
Breast MRI is recommended by the American College of Radiology Appropriateness Criteria® as the preferred modality for assessing treatment response to NAC [5]. Diffusion weighted imaging (DWI) provides a quantitative assessment of tissue microstructure by measuring water molecule diffusion, and its derived apparent diffusion coefficient (ADC) has been widely investigated as a biomarker of treatment response [6–8]. Higher pretreatment ADC values in patients with early-stage breast cancer have been reported to be associated with an increased likelihood of achieving pCR [9]. However, the broader implementation of MRI is limited by its relatively long acquisition times, high cost, and contraindications in patients with claustrophobia or certain implanted medical devices.
Given these constraints, there is growing interest in exploring more accessible and versatile imaging alternatives. Contrast-enhanced CT of the chest is routinely employed in the staging and surveillance of breast cancer, particularly for evaluating thoracic metastases (2). Given routine thoracic coverage, CT inherently captures breast tissue, presenting an opportunistic platform for assessing the primary breast tumor without additional radiation exposure.
Dual-energy CT (DECT) offers superior image quality and additional information over conventional CT by generating functional quantitative parameters such as normalized iodine concentration (NIC) and effective atomic number (Zeff) [10]. These parameters have demonstrated potential in breast lesion detection, tumor characterization, and treatment response assessment [11–13]. Furthermore, longitudinal changes in NIC have been established as an indicator of treatment efficacy following NAC in gastric cancer [14, 15]. Despite these advances, the comparative performance between DECT-derived quantitative parameters and ADC values from DWI remains inadequately explored in the context of breast cancer.
Therefore, we hypothesized that quantitative parameters from DECT could serve as early predictors of pCR. This study aims to evaluate the feasibility of DECT parameters for predicting pCR to NAC in breast cancer and to compare their predictive performance against ADC values obtained from DWI.

Methods

Participants
This prospective study complied with the principles of the Declaration of Helsinki and received approval from the institutional review board of our hospital. Written informed consent was obtained from all enrolled participants. Between June 2022 and December 2023, participants with histopathologically confirmed invasive breast cancer who underwent pretreatment breast MRI and chest DECT as part of the clinical staging work-up were prospectively enrolled. Inclusion criteria were as follows: (i) completion of both DECT and MRI (including DWI) before NAC; (ii) presence of a breast mass > 1 cm in diameter on contrast-enhanced MRI; (iii) no prior anti-tumor treatment and completion of the entire NAC regimen; and (iv) subsequent surgery with pCR assessment based on postoperative histopathology. Exclusion criteria included: (i) incomplete clinicopathologic data (n = 17); (ii) breast biopsy performed within one week before DECT examination (n = 36); (iii) lesions not fully covered within the imaging field of view due to large breast size (n = 4), and (iv) poor image quality on DECT or ADC maps (n = 19). Finally, 65 patients were included. The flowchart is summarized in Fig. 1.

DECT acquisition protocol
CT images were acquired using a 2.5-generation dual-source CT scanner (SOMATOM Drive, Siemens Healthineers, Forchheim, Germany) in dual-energy mode with two X-ray tubes operating at different voltages (tube A, 100 kVp; tube B, Sn140 kVp with a tin filter for high-voltage optimization). Automatic exposure control (CARE Dose 4D, Siemens Healthineers) was applied to all examinations. Patients were scanned craniocaudally in the supine position with both arms raised alongside the head to minimize beam-hardening artifacts. Scan parameters were as follows: collimation, 64 × 0.6 mm; rotation time, 0.28 s; pitch, 0.55; and reference tube current-time product of 71 mAs at 100-kVp tube and 60 mAs at Sn140-kVp. Images were reconstructed with both section thickness and an increment of 1.5 mm. For contrast-enhanced scanning, a nonionic iodinated contrast agent (Ioversol, 320 mg iodine/mL, Hengrui®, Jiangsu, China) was administered via the right or left ulnar vein using a dual-head injector at a dose of 1.5 mL/kg and a flow rate of 2.5 mL/s, followed by a 30-mL saline flush at the same rate. The arterial-phase imaging was triggered using a bolus-tracking technique with a threshold of 100 Hounsfield units (HU) in the descending aorta and an additional 10-sec delay. The subsequent venous-phase imaging was obtained 25 s after the completion of the arterial phase, which corresponds to approximately 60–90 s after injection start in most patients, depending on individual bolus arrival and scan length. Therefore, the venous-phase iodine metrics in this study should be interpreted as reflecting near-peak enhancement, rather than a wash-out phase. A dedicated late delayed phase was not acquired to avoid additional radiation exposure. This biphasic protocol has been consistently employed in prior studies investigating DECT for breast lesion characterization [12, 16].

MRI acquisition protocol
MRI was performed on a 3.0-T system (Ingenia, Philips Healthcare, the Netherlands) with an eight-channel breast array coil. The breast MRI acquisition protocol included axial T1-weighted, T2-weighted with fat suppression, DWI, and DCE-MRI sequences. DWI was performed by using a single-shot echo planar imaging sequence with parallel imaging (repetition time > 4000 msec; echo time, minimum; field of view, 300–360 mm; flip angle, 90°; acquired matrix, 128 × 128 to 192 × 192; section thickness, 4–5 mm; and acquisition time, ≤ 5 min). Diffusion gradients were applied in three orthogonal directions with b values of 0 and 800 s/mm2. The detailed MRI protocol is provided in Table S1 (supplementary material).

Image interpretation
DECT data were analyzed using the viewer software on a syngo.via workstation (VB20A, Dual Energy, Siemens Healthineers, Forchheim, Germany). Standard linear-blended images were reconstructed by applying a blending factor of 0.5 (M_0.5; 50% low kV and 50% high kV spectra) to obtain attenuation (HU) measurements. Two radiologists (J.F.S. and L.H.D, with 10 years and 5 years of experience in breast and chest diagnostic imaging, respectively) independently performed the measurements.
DECT parameters were measured by placing a region of interest (ROI) on the maximal axial section of each tumor while avoiding areas of calcification, necrosis, cysts, and vessels. ROI measurements were repeated with an interval of at least 1 month by the senior radiologist to assess intra-observer agreement. Quantitative parameters, including CT values measured on monochromatic images, the iodine concentration derived from decomposition images of iodine-based material, and the effective atomic number, were calculated using workstation software.
The normalized iodine concentration (NIC) and normalized effective atomic number (nZeff) were obtained by dividing the iodine concentration (mg/cm3) and effective atomic number of the lesion by the iodine concentration and effective atomic number of the aorta, respectively. The slope of the spectral HU curve (λHu, in HU per kiloelectron volt [keV]) was defined as the difference between the CT value at 40 keV and that at 70 keV divided by the energy difference (30 keV), calculated as:

The difference between arterial-phase NIC (A-NIC) and venous-phase NIC (V-NIC) was defined as:

The difference between arterial-phase nZeff (A-nZeff) and venous-phase nZeff (V-nZeff) was defined as:

The difference between arterial phase λHu (A-λHu) and venous phase nZeff (V-λHu) was defined as:

For the quantitative measurements on MRI, breast lesions were first identified on subtraction DCE-MRI images. Subsequently, ADC values were measured by manually delineating ROIs on ADC maps at the slice with the largest tumor diameter, using high–b-value DWI images (b = 800 s/mm2) as a reference to ensure accurate localization of diffusion-restricted regions while excluding adjacent adipose or fibroglandular tissue.
Both readers were blinded to clinicopathologic information and pCR status. For all subsequent statistical analyses, measurements obtained by the more experienced radiologist (J.F.S.) were used as the primary dataset; measurements from the second radiologist (L.H.D.) were used only for inter-observer agreement assessment. For intra-observer agreement, the more experienced radiologist repeated all measurements after a 3-month interval, blinded to the initial results.
Additionally, a qualitative assessment of pretreatment MRI was performed to account for the influence of tissue composition on ADC measurements. The diffusion-related imaging features included (1) intratumoral necrosis/cystic change, defined as a focal non-enhancing area on DCE subtraction images with corresponding high signal on T2-weighted fat-suppressed images; (2) intratumoral T2 hyperintensity, defined as visually apparent high signal within the lesion on T2-weighted fat-suppressed images; and (3) peritumoral edema, defined as T2-weighted fat-suppressed hyperintensity in the peritumoral breast tissue. Each feature was recorded as present or absent.

NAC regimens and pathologic response evaluation
All participants received 4–8 cycles of anthracycline-based, taxane-based, or combined regimens of taxanes and anthracyclines following the National Comprehensive Cancer Network (NCCN) Guidelines® [17]. Participants with human epidermal growth factor receptor-2 (HER2) positive breast cancer additionally received trastuzumab and/or pertuzumab. The status of estrogen receptor (ER), progesterone receptor (PR), HER2, and Ki67 index was assessed by immunohistochemistry with core needle biopsy specimens before NAC (Appendix S1, supplemental material). Treatment response was classified as pCR or non-pCR based on surgical resection specimens after neoadjuvant chemotherapy. pCR was defined as no residual invasive carcinoma in the breast (allowing for residual ductal carcinoma in situ) and no evidence of axillary lymph node involvement (ypT0/is ypN0).

Statistical analysis
Statistical analyses were performed using SPSS software (version 27; IBM Corp.). Inter-observer and intra-observer reproducibility of quantitative DECT parameters was assessed using intraclass correlation coefficients (ICCs) with a two-way random-effects model of consistency (ICC > 0.75 indicating excellent agreement; 0.60–0.74, good; 0.40–0.59, moderate; and < 0.40, poor). Continuous variables were compared using the Student t-test for normally distributed data and the Mann-Whitney U test for non-normally distributed data. Categorical variables were compared using the Chi-square test or Fisher exact test when expected cell counts were < 5. Correlations between quantitative DECT parameters and tumor size were assessed using Spearman’s rank correlation coefficients (ρ). Receiver operating characteristic (ROC) curve analysis was performed to assess the performance of DECT quantitative parameters and ADC value for predicting pCR. Sensitivity, specificity, and accuracy were calculated, and optimal thresholds were determined using the Youden index. The Delong test was used to compare the areas under the ROC curves (AUCs). Statistical significance was defined as a two-sided p value < 0.05.

Results

Participant characteristics
The demographic and clinicopathologic characteristics of the participants are summarized in Table 1. A total of 65 women (mean age ± SD, 51 ± 7 years) were included. Of these, eight participants (12%) had luminal A breast cancer, 31 participants (48%) had luminal B breast cancer, 14 participants (22%) had triple-negative breast cancer, and 12 participants (18%) had HER2-enriched breast cancer. After NAC, 15 participants (23%) achieved pCR, while 50 participants (77%) did not.
Participants who achieved pCR were more likely to have a T1/T2 clinical tumor stage (cT) than a T3/T4 stage and to have clinical stage IIA/IIB versus stage IIIA/IIIB/IIIC breast cancer (p = 0.002 and 0.005, respectively). No evidence of significant differences was observed between the pCR and non-pCR groups regarding age, menopausal status, clinical nodal stage, ER/PR status, HER2 status, Ki67, molecular subtype, or tumor size.
We additionally performed a qualitative review of pretreatment MRI for diffusivity-related features, including intratumoral necrosis/cystic change, intratumoral T2 high signal intensity, and peritumoral edema. The prevalence of these features did not differ between participants with and without pCR (p = 0.89, 0.76, and 0.48, respectively; Table S3).

Comparison of DECT parameters and ADC between pCR and non-pCR groups
Table 2 displays the DECT parameters and ADC in participants with and without pCR. Venous-phase NIC (V-NIC) (0.29 ± 0.10 vs. 0.39 ± 0.13, p = 0.004) and ∆NIC (0.20 ± 0.08 vs. 0.26 ± 0.09, p = 0.03) were lower in the pCR group than in the non-pCR group. The ADC was higher in the pCR group than in the non-pCR group (0.89 ± 0.13 vs. 0.77 ± 0.13 × 10⁻³ mm²/s, p = 0.004) (Fig. 2). Other DECT parameters, including arterial-phase NIC (A-NIC), nZeff, λHu, and CT attenuation, showed no statistical differences between the pCR and non-pCR groups (all p > 0.05). Representative DECT and diffusion-weighted MRI images are shown in Fig. 3.

Predictive performance of quantitative DECT parameters and ADC
ROC curve analysis (Table 3; Fig. 4) demonstrated that V-NIC yielded an AUC of 0.75 (95% confidence interval [CI]: 0.62, 0.85). Using a cutoff value of 0.32, V-NIC achieved a sensitivity of 73% (11/15), specificity of 70% (35/50), and accuracy of 71% (46/65). Combining V-NIC and ∆NIC produced an AUC of 0.75 (95% CI: 0.62, 0.85) with the same sensitivity, specificity, and accuracy as V-NIC alone (p = 0.73). The ADC achieved an AUC of 0.71 (95% CI: 0.58, 0.81), with a sensitivity of 87% (13/15), specificity of 50% (25/50), and accuracy of 58% (38/65). The combination of quantitative DECT parameters (V-NIC or V-NIC + ΔNIC) with ADC yielded an AUC of 0.79 (95% CI: 0.65–0.90) for predicting pCR. Both models showed 100% sensitivity (15/15), though specificity and accuracy were moderate: 48% (24/50) and 60% (39/65) for V-NIC + ADC, and 46% (23/50) and 59% (38/65) for V-NIC + ΔNIC + ADC, respectively.
The difference in AUC between V-NIC and ADC was not statistically significant (p = 0.69). Similarly, no significant differences were observed between the combined parameters (V-NIC + ADC and V-NIC + ΔNIC + ADC) and the individual parameters V-NIC (p = 0.31 and 0.34, respectively) or ADC (P = 0.18 and 0.20, respectively) (Table S2 Supplementary material).

Incremental value beyond clinical staging and assessment of potential confounding by tumor burden
Given the imbalance in cT stage and overall clinical stage between the pCR and non-pCR groups, we examined whether V-NIC might act as a surrogate for tumor burden or stage. V-NIC was not significantly correlated with tumor size (Spearman ρ = 0.13, p = 0.29). In subgroup analyses, V-NIC did not differ significantly between cT1–2 and cT3–4 tumors (p = 0.20; Table S3) or between clinical stage IIA–IIB and IIIA–IIIC tumors (p = 0.40; Table S4). We further explored the incremental predictive value of V-NIC beyond the cT stage. The model combining cT stage and V-NIC yielded a higher AUC (0.88 [95% CI: 0.75–0.94]) than cT stage alone (0.73 [95% CI: 0.61–0.84]; p = 0.01) (Table 3 and S2).

Intra- and inter-observer agreement for DECT parameters and ADC
According to the ICC, intra- and inter-observer agreement for all DECT parameters and ADC were all good to excellent (ICC range: 0.65–0.98), as shown in Table 4.

Discussion
In this study, we investigated and compared the performance of DECT and DWI for predicting pCR after NAC for breast cancer. We found that the V-NIC was lower and ADC was higher in participants who achieved pCR than in those who did not (both p = 0.004). The AUC of V-NIC was not significantly different from that of ADC (0.75 vs. 0.71, p = 0.69). Additionally, the combination of V-NIC and ADC (AUC = 0.79) did not improve the predictive performance compared with V-NIC or ADC alone (p = 0.31 and 0.18, respectively). Because the cT stage differed between the pCR and non-pCR groups, we assessed whether V-NIC might simply serve as a surrogate of tumor burden. V-NIC showed no significant correlation with tumor size (Spearman ρ = 0.13, p = 0.29) and did not differ across cT or clinical stage categories (p = 0.20 and 0.40, respectively). Moreover, adding V-NIC to a cT-based model significantly improved predictive performance (0.88 vs. 0.73, p = 0.01), suggesting incremental value of DECT beyond clinical staging. Given the modest sample size, these findings should not be interpreted as evidence of equivalence but rather as preliminary evidence of the feasibility of pretreatment DECT for predicting response to NAC in breast cancer and require validation in larger, independent cohorts.
In our cohort, pretreatment V-NIC was lower in participants who achieved pCR. Given that V-NIC was derived from a single early venous-phase acquisition, the measured iodine signal should be interpreted as reflecting the lesion iodine distribution at a specific time point rather than a direct surrogate of efficient perfusion [12]. DECT iodine metrics have been shown to reflect contrast distribution and to correlate with vascular-related properties such as blood volume and microvessel density [18–20]. However, at this timing the measured iodine signal is plausibly influenced by both intravascular contrast and early extravascular distribution driven by permeability and interstitial transport. Tumor microvessels are frequently structurally and functionally abnormal; increased permeability and vascular disorganization can elevate interstitial fluid pressure, exacerbate heterogeneous perfusion, and reduce transvascular transport, ultimately limiting intratumoral drug penetration despite apparent enhancement on imaging [21]. Under this framework, a higher early venous-phase iodine signal may partially reflect permeability-related extravasation and early interstitial retention, whereas lower V-NIC may be more consistent with less iodine retention and relatively more efficient microcirculatory function, which could align with the higher pCR rate observed in our dataset. These mechanistic hypotheses should be tested in future studies using dynamic multi-phase or perfusion imaging.
The better predictive performance of the venous-phase over arterial-phase NIC in our data aligns with prior work and with contrast-kinetic principles [16, 22]. Arterial-phase iodine is sensitive to bolus timing, cardiac output, and macrovascular inflow, which inflate inter-subject variability. Moreover, the DECT protocol used in this study was adapted from a thoracic contrast-enhanced CT protocol, which may have limited contrast permeation into breast lesions during the arterial phase, making venous-phase imaging more informative for assessing intralesional microvessel density [16]. Importantly, NIC, rather than IC, mitigates inter-individual hemodynamic confounders by normalizing lesion iodine to a reference vessel, improving reproducibility when comparing patients or scanners [23]. Prior breast and gastric cancer studies recommend standardized iodine metrics to reduce systemic circulation effects and to more faithfully reflect vascularity features and iodine deposition, consistent with our approach [24, 25].
Regarding diffusion-weighted imaging, we observed significantly higher pretreatment ADC values in participants who achieved pCR than in those who did not. However, the evidence supporting pretreatment ADC as an independent predictor of pCR remains inconsistent, with substantial heterogeneity across cohorts, acquisition parameters, and analysis strategies [6, 26–28]. To further contextualize our baseline ADC finding, we assessed pretreatment MRI features that could potentially increase diffusivity, including intratumoral necrosis or cystic change, intratumoral T2 hyperintensity, and peritumoral edema; none significantly differed between the pCR and non-pCR groups (all p > 0.05). These negative results suggest that gross, visually apparent necrosis or edema alone is unlikely to account for the baseline ADC pattern observed in our cohort. Nevertheless, more subtle contributors, such as microscopic necrosis, edema, stromal composition, or ROI-related partial-volume effects, may not be captured by a binary morphologic score and could still affect ADC measurements. Given the moderate inter-observer agreement for ADC, we performed a sensitivity analysis using the second radiologist’s measurements. The pCR group still showed significantly higher ADC than the non-pCR group (0.91 ± 0.14 vs. 0.78 ± 0.17; p = 0.01), indicating that the statistical finding was robust to reader-related variability. In addition, accumulating evidence has indicated that the longitudinal changes in ADC during NAC may be more informative of response than a single pretreatment ADC measurement [29, 30]. Collectively, larger studies with standardized diffusion protocols and longitudinal measurements are needed to clarify the independent predictive value of baseline ADC.
Beyond V-NIC and ADC, other DECT parameters, such as λHu and nZeff, did not exhibit value in predicting pCR in this study. Although λHu has been reported to help detect metastatic sentinel lymph nodes in breast cancer and nZeff can differentiate certain molecular subtypes [10, 31], neither metric appears to directly reflect microvascular density or perfusion, which are the key pathophysiologic determinants of NAC response. These negative findings highlight that not all DECT–derived parameters are equally informative for treatment response prediction.
This study has several limitations. First, the single-center design and modest sample size, with a limited number of pCR events, may constrain generalizability and preclude subgroup analyses based on molecular subtypes. Accordingly, the reported AUC and cut-off values should be regarded as preliminary and require validation in larger, independent cohorts before clinical implementation. Second, quantitative measurements were obtained from a two-dimensional ROI placed on the maximal axial slice rather than whole-tumor volumetric segmentation. Given the lesion size in this cohort, this approach may not fully capture intratumoral heterogeneity and may affect metric robustness, although it was selected to preserve clinical feasibility and reproducibility. Future studies incorporating whole-lesion analyses may help improve the representativeness of quantitative measurements. Third, ADC measurements showed only moderate inter-observer agreement, which may increase measurement error and inflate within-group variance, thereby attenuating effect estimates. Although the between-group difference was reproducible using an independent second-reader assessment, larger multicenter cohorts with potentially automated or semi-automated segmentation are warranted to further improve reproducibility and confirm generalizability. Finally, our analysis was limited to pretreatment examinations. The temporal changes of DECT parameters during NAC warrant investigation in future studies through the collection of longitudinal data at multiple timepoints.

Conclusions
In conclusion, this study provides preliminary evidence supporting the feasibility of pretreatment DECT as a noninvasive approach for predicting pCR after NAC in breast cancer, with predictive performance not significantly different from DWI. However, due to the limited sample size, these results should be interpreted with caution and require confirmation in larger, independent cohorts.

Supplementary Information
Below is the link to the electronic supplementary material.