ABSTRACT

Objective

To evaluate the predictive performance of dimensional and functional magnetic resonance imaging (MRI) parameters obtained from mid-treatment breast MRI for forecasting pathologic complete response (pCR) in patients with locally advanced breast cancer (LABC) undergoing neoadjuvant chemotherapy (NAC).

Materials and Methods

Sixty-five women with LABC who underwent NAC followed by surgery were retrospectively included. Quantitative MRI parameters—including % change (Δ%) in longest diameter, bidimensional size, tumor volume, apparent diffusion coefficient (ADC), and enhancement percentage (Epeak) —were calculated between pre- and mid-NAC MRI. Receiver operating characteristic (ROC) analysis and logistic regression were used to identify predictors of pCR. Logistic regression and ROC analysis (with DeLong’s test) were used to assess associations with pCR and compare area under the curves (AUCs).

Results

pCR was achieved in 19 of 65 patients (29%). Compared to non-pCR cases, patients with pCR showed significantly greater reductions in tumor size and Epeak, and larger increases in ADC value (all p<0.05). In multiple logistic regression, Δ% longest diameter >60% [odds ratio (OR)=7.1, p = 0.008] and Δ% ADC value ≥32% (OR=4.7, p = 0.016) remained statistically significant independent predictors of pCR. Δ% tumor volume >92% had the highest univariable AUC (0.754), while Epeak ≤21% showed perfect specificity but was excluded due to wide confidence intervals. Pairwise AUC comparisons showed no significant differences among Δ% longest diameter, bidimensional size, and tumor volume (all p>0.05).

Conclusion

Mid-treatment MRI biomarkers, particularly Δ% longest diameter and Δ% ADC value, are effective early predictors of pCR and may support individualized treatment strategies during NAC.

Introduction

Neoadjuvant chemotherapy (NAC) has become a cornerstone in the management of locally advanced breast cancer (LABC), enabling tumor downstaging, facilitating breast-conserving surgery, and offering an in vivo assessment of chemosensitivity prior to surgery. One of the critical surrogate endpoints used to evaluate NAC efficacy is the achievement of a pathologic complete response (pCR), which has been shown to correlate with improved long-term outcomes, particularly in HER2-positive and triple-negative subtypes (1, 2).

Early prediction of pCR is critical for guiding treatment decisions, including de-escalation or intensification of therapy. Magnetic resonance imaging (MRI) has been shown to outperform conventional modalities such as mammography, digital breast tomosynthesis, and automated breast ultrasound, while emerging techniques like contrast-enhanced spectral mammography and positron emission tomography/MRI offer additional functional insights (3-6).

Among the various MRI time points, mid-treatment (mid-NAC) MRI, typically performed after 2 to 4 cycles of chemotherapy, provides a unique opportunity to assess early response and adapt treatment plans in real time (7, 8). This time point allows clinicians to evaluate early response and modify treatment plans accordingly, potentially avoiding unnecessary toxicity or surgical delays.

Building on this potential, several studies have evaluated the predictive value of MRI-derived quantitative parameters in predicting pCR to NAC in breast cancer patients. These parameters include not only morphologic features such as tumor diameter and volume but also functional measures like peak enhancement percentage (Epeak) and apparent diffusion coefficient (ADC) (7-10).

Tudorica et al. (8) demonstrated that early changes in functional MRI parameters after just one cycle of NAC were more effective in predicting pCR than conventional size-based criteria. Similarly, in a prospective study on inflammatory breast cancer, Le-Petross et al. (11) reported that mid-treatment median and mean ADC values were significantly associated with pCR, highlighting their potential as early imaging biomarkers. Despite these encouraging findings, the overall results remain heterogeneous, and there is still no consensus regarding the most reliable parameters or optimal cut-off values. Moreover, although mid-NAC MRI has attracted growing interest, the majority of existing research has focused on post-treatment imaging. Therefore, further validation in well-defined patient cohorts is needed to clarify the combined predictive value of morphologic and functional MRI biomarkers at the mid-NAC stage (4, 8).

Therefore, this study aims to evaluate the predictive performance of both morphologic and functional MRI biomarkers obtained at the mid-treatment stage for forecasting pCR in patients with LABC receiving NAC. By investigating multiple MRI-derived parameters, including tumor size, Epeak, and ADC, we seek to contribute to the development of reliable imaging-based predictors that can guide early treatment decisions.

Materials and Methods

Study Design and Patient Selection

This retrospective study was approved by the University of Health Sciences Türkiye, İzmir Bozyaka Training and Research Hospital Clinical Research Ethics Committee (approval date: 29/11/2023; decision no: 2023/197). Patients diagnosed with breast cancer at our institution between 2019 and 2021 were evaluated for eligibility. Inclusion criteria were as follows: receipt of NAC; availability of breast MRI at three distinct time points—prior to NAC (pre-NAC), mid-treatment (after the fourth chemotherapy cycle) (mid-NAC), and post-treatment (post-NAC); and subsequent surgery performed at our institution with pathological assessment of treatment response. Of the initially screened patients, those were excluded due to incomplete or prematurely terminated NAC or non-adherence to the treatment protocol (n = 5), lack of pre-treatment pathological confirmation at our institution (n = 2), absence of appropriately timed or technically adequate MRI examinations (n = 4), history of prior malignancy or cancer treatment (n = 1), or surgery performed outside our institution without pathological response evaluation (n = 2). After applying these criteria, a total of 65 patients were included in the final analysis.

Chemotherapy Regimens

All patients completed the prescribed systemic NAC and subsequently underwent surgery. NAC regimens were categorized into three distinct groups: anthracycline-based regimens (doxorubicin or epirubicin), taxane-based regimens (docetaxel or paclitaxel), and sequential regimens combining both anthracyclines and taxanes (Table 1). No patients received neoadjuvant endocrine therapy or radiotherapy.

Histopathological Evaluation

Breast tumor samples obtained through tru-cut biopsy were classified into three molecular subgroups based on immunohistochemical markers: hormone receptor–positive [estrogen receptor (ER) and/or progesterone receptor (PR) positive, human epidermal growth factor receptor (HER2)-negative], HER2-positive (regardless of hormone receptor status), and triple-negative (ER-negative, PR-negative, and HER2-negative). A tumor was considered hormone receptor–positive if either ER or PR showed ≥1% positivity on immunohistochemistry.

Following surgery, the response to NAC was assessed using the Miller–Payne grading system (Grade 1–5) (12). Grades 1 to 4 were classified as non-pCR, while Grade 5 was defined as pCR. The presence of ductal carcinoma in situ was excluded from the definition of pCR. Similarly, axillary nodal status was not considered in the pCR assessment.

MRI Technique

All breast MRI examinations were performed using a 1.5 Tesla scanner (Magnetom Aera, Siemens Healthineers, Erlangen, Germany) equipped with a dedicated breast coil. Imaging was conducted at three predefined time points: pre-NAC, mid-NAC (after the fourth chemotherapy cycle), and post-NAC. Mid-NAC MRI was performed at a mean of 8.4±1.8 weeks after the initiation of chemotherapy, with a median of 8.3 weeks [interquartile range (IQR): 7.3–9.5 weeks].

The standardized protocol included the following sequences in the axial plane: T1-weighted fat-saturated (TR/TE: 476/11 ms, slice thickness: 4.0 mm), T2-weighted turbo spin-echo (TR/TE: 6240/76 ms, slice thickness: 4.0 mm), turbo inversion recovery magnitude (TR/TE: 2250/56 ms), and T1-weighted Dixon (TR/TE: 449/11 ms). Diffusion-weighted imaging (DWI) was acquired using a single-shot echo-planar imaging sequence with b-values of 50, 500, and 800 s/mm² (TR/TE: 6900/66 ms, slice thickness: 4.0 mm). Dynamic contrast-enhanced (DCE) MRI was obtained using a 3D fat-saturated T1-weighted gradient-echo sequence (TR/TE: 4.53/1.82 ms, slice thickness: 2.0 mm). A gadolinium-based contrast agent (gadobutrol, Gadovist^®; Bayer, Berlin, Germany) was administered intravenously at a dose of 0.1 mmol/kg, followed by a 20 mL saline flush at a rate of 3 mL/s. Five sequential post-contrast phases were acquired, each lasting 90 seconds, with the first acquisition centering k-space at approximately 1.5 minutes after injection.

All participants were provided informed consent before undergoing imaging.

Radiologic Response Evaluation

All MRI studies were reviewed in consensus by two radiologists with 8 and 23 years of experience, using dedicated workstations (Siemens Healthineers), blinded to the pathological outcomes. Radiological response to NAC was assessed according to RECIST 1.1 criteria (13). For multifocal or multicentric breast cancer, up to two target lesions—preferably the largest—were selected for serial measurements. Non-target lesions were qualitatively assessed but not quantitatively measured.

Radiologic complete response (rCR) was defined as the complete disappearance of all measurable lesions without any residual contrast enhancement. If a residual structure was observed at the tumor site, additional evaluation with DWI was performed. Lesions lacking contrast enhancement were still classified as complete response. Patients who did not fulfill these criteria were categorized as non-rCR.

For each target lesion, the longest diameter, the second-longest diameter, and the shortest diameter were recorded. Based on these dimensions, bidimensional size and tumor volume were calculated. Bidimensional size (cm²) was defined as the product of the longest and second-longest diameters. Tumor volume (cm³) was estimated using the ellipsoid formula:

Volume = (π/6) × longest diameter × second diameter × shortest diameter

Signal intensity (SI) measurements were obtained on both pre-NAC and mid-NAC DCE-MRI scans. Two key values were recorded: The baseline SI before contrast injection (SI_pre) and the peak signal intensity during the early post-contrast phase, approximately two minutes after injection (SI_peak) (10, 14). The early peak enhancement (Epeak) was calculated using the formula:

Epeak (%) = [(SI_peak − SI_pre) / SI_pre] × 100

In patients with rCR, SI measurements on mid-NAC MRI were obtained from the anatomical location corresponding to the original tumor site.

ADC values were measured by placing circular regions of interest (ROIs) within the most diffusion-restricted area of each lesion, identified on high b-value DWI. ROIs were placed on the corresponding hypointense region on the ADC map to accurately reflect the most cellular portion of the tumor. Each ROI had a standardized area of 0.5 cm². Necrotic or cystic areas were carefully avoided, as they could artificially elevate ADC values (Figure 1). In rCR cases, ADC values were measured from the original tumor site on mid-NAC MRI. For patients with multifocal disease, the largest lesion was selected as the primary target for quantitative analysis.

All quantitative measurements were performed once per lesion without repetition.

Percentage changes (Δ%) in tumor dimensions and quantitative MRI parameters between pre-NAC and mid-NAC were calculated using the formula:

Δ% = [(Pre − Mid) / Pre] × 100

Statistical Analysis

All statistical analyses were performed using SPSS version 26.0 (IBM Corp., Armonk, NY, USA). A two-tailed p-value of <0.05 was considered statistically significant. Categorical variables, including clinical and pathological characteristics, were summarized as frequencies and percentages. Continuous variables, such as tumor dimensions and MRI-derived parameters, were presented as mean ± standard deviation, median (minimum–maximum), and IQR (25^th–75^th percentiles). The Mann-Whitney U test was used to compare non-normally distributed variables between groups. Diagnostic performance was evaluated using receiver operating characteristic (ROC) curve analysis, and optimal cut-off values were determined using Youden’s index. These thresholds were used to derive dichotomous variables for further analysis. Associations between radiologic parameters and pCR were assessed using Yates’ corrected chi-square test or Fisher’s exact test, depending on expected frequencies. Univariate logistic regression was used to calculate odds ratios (ORs) with corresponding 95% confidence intervals (CIs). Variables with statistical significance in univariate analysis were included in a multiple logistic regression (MLR) model using the backward stepwise (Wald) method. Model calibration was evaluated using the Hosmer-Lemeshow goodness-of-fit test, and multicollinearity among predictors was assessed to ensure the retention of independent variables. The assumption of normality was tested using the Shapiro-Wilk test and visual inspection of Q-Q plots. Pairwise comparisons of area under the curves (AUC) values between MRI parameters were performed using DeLong’s test, a non-parametric method for comparing correlated ROC curves.

Results

The study cohort included 65 female breast cancer patients, with a mean age at diagnosis of 50±9 years (IQR: 44–56). pCR of the primary tumor was achieved in 19 patients (29%), while 46 patients (71%) were classified as non-pCR (Table 1).

Quantitative MRI analysis showed a marked reduction in tumor dimensions and enhancement characteristics from pre- to mid-NAC. The mean longest diameter decreased from 3.47±1.77 cm to 2.08±1.48 cm, with a mean percent change (Δ%) of 40%. Tumor volume decreased from 17.6±36.3 cm³ to 5.4±17.7 cm³ (Δ%: 71%). Epeak declined from 177±101% to 100±79% (Δ%: 23%), while ADC values increased from 867±180 to 1035±148 ×10^–⁶ mm²/s (Δ%: -23%) (Table 2). Notably, due to the calculation formula [(Pre − Mid) / Pre × 100], negative Δ% ADC values indicate an actual increase in ADC.

Patients who achieved pCR exhibited significantly greater reductions in tumor size and enhancement metrics on mid-NAC MRI. The mean percent decrease (Δ%) in longest diameter, second diameter, shortest diameter, tumor volume, and bidimensional size was significantly higher in the pCR group compared to non-pCR (all p<0.01) (Table 3). Notably, mid-NAC ADC values were significantly higher (1107 vs. 1006 ×10^–⁶ mm²/s, p = 0.015), and mid-NAC Epeak values were lower (68 vs. 113%, p = 0.007) in patients with pCR.

ROC analysis demonstrated that several mid-NAC MRI-based parameters had good discriminatory ability for predicting pCR (Table 4, Figure 2). Δ% longest diameter >60% achieved an AUC of 0.740 (95% CI: 0.616–0.841) with high specificity (91%). Δ% second diameter >46% provided the highest sensitivity (74%) with 70% specificity (AUC=0.752; 95% CI: 0.630–0.851). Tumor volume reduction >92% also showed strong performance with an AUC of 0.754 (95% CI: 0.631–0.852) and specificity of 89%. Among functional metrics, Epeak ≤21% yielded perfect specificity (100%) with an AUC of 0.715 (95% CI: 0.590–0.820), although sensitivity was lower (37%). Most dimensional parameters achieved AUC values >0.73 (p<0.01), supporting their predictive utility (Table 4).

Pairwise comparisons of AUC values between Δ% MRI parameters—including diameters, bidimensional size, tumor volume, and ADC—showed no significant differences (all p>0.05). Δ% dimensional metrics demonstrated highly similar performance (AUC differences <0.015), while Δ% ADC tended to yield slightly higher AUCs without reaching significance (Supplementary Table 1).

Binary analysis based on optimal cut-off values confirmed significant associations between mid-NAC MRI–derived parameters and pCR (Table 5). Patients with Δ% longest diameter >60%, second diameter >46%, shortest diameter >45%, or tumor volume reduction >92% had significantly higher rates of pCR (all p<0.01). Additionally, patients with Epeak ≤21% and Δ% ADC value ≥32% showed significantly higher pCR rates (p<0.01 for both).

In univariable analysis, multiple mid-NAC MRI–based parameters were significantly associated with pCR, including Δ% longest diameter >60% (OR=9.5; 95% CI: 2.4–37.0; p = 0.001), Δ% ADC value ≥32% (OR=6.2; 95% CI: 1.9–19.8; p = 0.003), and Δ% tumor volume >92% (OR=5.3; 95% CI: 1.6–17.2; p = 0.010) (Table 6). Epeak ≤21% demonstrated the highest univariable OR (OR=26.3; 95% CI: 2.9–234.5; p<0.001) but was excluded from the MLR model due to wide CIs indicating model instability. Δ% second diameter, shortest diameter, and bidimensional size also showed significant associations (all p<0.01) but were excluded due to collinearity.

In the final MLR model, only Δ% longest diameter >60% (OR=7.1; 95% CI: 1.3–30.1; p = 0.008) and Δ% ADC value ≥32% (OR=4.7; 95% CI: 1.3–16.5; p = 0.016) remained as independent predictors of pCR (Table 6). The model excluded redundant or unstable variables, favoring robust and non-collinear biomarkers. Overall model performance was acceptable (Nagelkerke R² = 0.336), with good calibration (Hosmer-Lemeshow p = 0.800).

Discussion and Conclusion

In the overall cohort, all dimensional MRI parameters showed a mean reduction from pre- to mid-NAC, while mean post-contrast SI, Epeak, decreased and mean ADC values increased, consistent with early treatment response (Table 2). Furthermore, comparison between pCR and non-pCR groups revealed significantly greater reductions in longest diameter, second and shortest diameters, bidimensional size, and tumor volume, as well as lower mid-treatment Epeak and higher ADC values in the pCR cohort (all p<0.05, Table 3). These differences underscore the utility of both morphologic and functional MRI parameters in distinguishing responders from non-responders during mid-NAC assessment, in line with prior studies that demonstrated the predictive value of early volumetric reduction and diffusion changes for pCR (7, 8, 11).

As shown in Table 3, pre-contrast SI on mid-NAC MRI exhibited a slight increase in the pCR group, whereas it remained stable or decreased in the non-pCR group; however, this difference did not reach statistical significance. This observation may be explained by treatment-induced stromal alterations such as proteinaceous or hemorrhagic content, fibrosis, and necrotic debris, which shorten T1 relaxation time and elevate signal intensity despite fat suppression. Nevertheless, given its limited clinical relevance, this parameter should not be considered a primary predictor of therapeutic response.

Univariable ROC analysis revealed that several morphologic and functional MRI parameters demonstrated significant predictive performance for pCR. Among dimensional metrics, tumor volume reduction >92% achieved the highest AUC (0.754), followed closely by bidimensional size >69% (AUC=0.752) and longest diameter >60% (AUC=0.740). Notably, the >60% reduction in the longest diameter yielded the highest specificity (91%), suggesting its utility in identifying true responders, whereas bidimensional shrinkage provided the highest sensitivity (74%), making it more suitable for screening purposes. These findings are in agreement with prior work by Fangberget et al. (15), who demonstrated that tumor volume reduction ≥83% after four cycles of NAC was the most accurate predictor of pCR, with a sensitivity of 91%, specificity of 80%, and AUC of 0.82, outperforming longest diameter reduction (AUC=0.78). This supports the notion that while unidimensional measures like longest diameter remain clinically useful, volumetric assessment better reflects the complex and often irregular patterns of tumor regression during NAC. Conversely, Minarikova et al. (16) reported that 3D diameter change outperformed volume metrics (AUC=0.933), suggesting that diameter-based measurements may offer better practicality and performance in certain clinical settings. In our cohort, however, pairwise AUC comparisons showed no statistically significant differences between Δ% longest diameter, bidirectional size, and tumor volume (all p>0.05), indicating comparable predictive performance across these morphologic parameters.

In the MLR analysis, Δ% longest diameter >60% (OR=7.1, p = 0.008) and Δ% ADC value ≥32% (OR = 4.7, p = 0.016) emerged as the only independent predictors of pCR. These findings reinforce the complementary value of morphologic shrinkage and diffusion metrics in early treatment monitoring. While other dimensional parameters also showed strong univariable associations with pCR, they were excluded from the final model due to collinearity. Notably, Δ% tumor volume >92%—despite having the highest univariable AUC (0.754)—lost significance in MLR analysis (p = 0.869), likely reflecting overlapping predictive information with longest diameter and ADC changes. This emphasizes the importance of selecting non-collinear and stable parameters for robust modeling in clinical practice.

The increase in ADC values, reflected by Δ% ADC ≥32%, suggests reduced tumor cellularity in response to effective cytotoxic therapy. Chemotherapy-induced necrosis and decreased cell density enhance water diffusivity, which is captured by ADC elevation on DW-MRI (17, 18). In our cohort, both higher mid-NAC ADC values and greater percent increases were significantly associated with pCR, consistent with prior findings from both multicenter and single-institution studies, supporting ADC as a reliable non-invasive biomarker of treatment response (7, 11). In contrast to prior multicenter studies such as ACRIN 6698, which demonstrated moderate predictive performance for mid-treatment ΔADC without specifying an optimal cut-off (AUC=0.60), our study identified a Δ% ADC ≥32% as an independent predictor of pCR with higher specificity and multivariable significance (7). Similarly, a recent multi-institutional study on HER2-positive breast cancer also reported ΔADC as a significant early predictor of treatment response (19).

Similarly, lower mid-NAC Epeak values—particularly those ≤21%—were highly specific for pCR (100%), although their predictive utility was limited by low sensitivity (37%) and instability in MLR modeling. As a semi-quantitative indicator of early contrast enhancement kinetics, Epeak reflects tumor vascularity and perfusion. A pronounced decline in Epeak following chemotherapy may correspond to reduced angiogenic activity, vessel permeability, and microvascular density, consistent with a strong cytotoxic effect. In our cohort, both lower absolute mid-NAC Epeak values and greater percent reductions were significantly associated with pCR in univariable analysis. These findings are in line with previous studies suggesting that early changes in perfusion-related parameters may signal effective treatment response. Supporting this, one recent study demonstrated that greater reductions in early peak enhancement during NAC were associated with favorable pathological response and improved recurrence-free survival (20). Additionally, a large retrospective cohort of 168 breast cancer patients reported that both pre-NAC Epeak ≤96 and post-NAC Epeak >188 were independent predictors of pCR and survival outcomes, further underscoring the prognostic value of contrast enhancement dynamics during treatment (10).

From a clinical perspective, the integration of early mid-treatment MRI biomarkers—particularly Δ% longest diameter and Δ% ADC—offers a practical and non-invasive strategy to identify patients who are likely to achieve pCR during neoadjuvant chemotherapy. These parameters can be easily derived from routine clinical MRI protocols without the need for advanced post-processing or pharmacokinetic modeling, increasing their feasibility in everyday practice. The high specificity of Δ% longest diameter >60% and the independent predictive value of Δ% ADC ≥32% support their utility in guiding individualized treatment decisions. For instance, patients demonstrating suboptimal early imaging response may be candidates for treatment intensification or alternative therapeutic strategies, while early identification of good responders may inform ongoing monitoring or risk stratification efforts. Thus, the use of accessible imaging biomarkers to differentiate responders from non-responders at an early stage may support more personalized and adaptive approaches in breast cancer management.

Study Limitations

This study has several limitations. First, its retrospective design and single-center setting may limit the generalizability of the findings. Second, the sample size—though adequate for exploratory modeling—was relatively modest, particularly for MLR, which may affect the stability of certain parameters, such as Epeak. Third, manual ROI placement for ADC and enhancement measurements, although performed in consensus by experienced radiologists, may introduce variability and operator dependence. Additionally, Epeak, as a semi-quantitative measure, is susceptible to temporal resolution and sequence timing, which could limit reproducibility across different imaging protocols. Moreover, the lack of inter-reader reproducibility testing further limits generalizability.

Mid-treatment MRI biomarkers—including Δ% longest diameter and Δ% ADC—demonstrated significant predictive value for pCR in patients undergoing NAC for LABC. These easily obtainable and non-invasive parameters may serve as reliable early indicators of treatment response, potentially enabling more personalized therapeutic strategies in clinical practice.