Comparative analysis of apparent diffusion coefficient (ADC) metrics for the differential diagnosis of breast mass lesions

Abstract

Background

Breast cancer’s diagnostic challenge is amplified by its heterogeneity. Diffusion-Weighted Imaging (DWI) offers promising avenues for precise tumor characterization through Apparent Diffusion Coefficient (ADC) metrics.

Purpose

To investigate the diagnostic utility of advanced ADC metrics in distinguishing breast lesions using Magnetic Resonance Imaging (MRI).

Methods

A retrospective cohort analysis of MRI data from 125 pathologically confirmed breast tumors was conducted. ADC values were independently measured by two physicians at the lesion sites and reference points (contralateral normal breast parenchyma, pectoralis major, and interventricular septum), from which advanced ADC metrics were calculated. Statistical analyses were applied to differentiate ADC metrics between malignant and benign groups. ROC curves assessed the diagnostic efficacy of individual ADC metrics. A binary logistic regression model incorporating ADC metrics and age was developed, with its diagnostic superiority evaluated through multidimensional comparisons.

Results

Of the 125 lesions, 77 were malignant and 48 benign. Significant differences in ADC metrics were found between malignant and benign tumors. Diagnostic analysis showed minimum ADC value (ADC_min) as the most effective single indicator, while the combined model, including age and average ADC value (ADC_avg), outperformed individual ADC metrics, demonstrating superior diagnostic accuracy (area under the curve (AUC) = 0.964). The combined model nomogram also showed improved clinical utility and a significant increase in diagnostic performance.

Conclusions

Advanced ADC metrics significantly enhance the diagnostic accuracy for differentiating between benign and malignant breast lesions. The development of a combined model further refines breast cancer diagnostics, supporting the advancement towards precision medicine.

Background

Breast cancer remains one of the leading causes of cancer incidence and mortality among women worldwide, with over 2 million new cases reported annually [1]. The early and accurate differentiation between benign and malignant breast tumors is critical for optimizing clinical management, minimizing unnecessary invasive procedures, and improving patient prognosis. Moreover, malignant breast tumors exhibit a range of pathological and molecular subtypes, each presenting distinct diagnostic and therapeutic challenges [2]. The inherent heterogeneity of these tumors further necessitates the advancement of diagnostic methods to accurately characterize and develop tailored treatment strategies. Conventional imaging techniques, including mammography and ultrasound, are widely employed for the initial characterization of breast lesions. However, these methods often have limitations in distinguishing between benign and malignant lesions, particularly in dense breast tissue or complex cystic-fixed masses [3]. In contrast, breast magnetic resonance imaging (MRI), particularly when combined with functional imaging techniques such as diffusion-weighted imaging (DWI), has emerged as an important tool for assessing tumor biology [4, 5].

DWI assesses the microscopic movement of water molecules within tissues, with the apparent diffusion coefficient (ADC) serving as a key quantitative metric derived from this technique [6]. This metric non-invasively reflects the cellular density and membrane integrity of tumor tissue, thus assisting in the differentiation of benign and malignant breast lesions [7]. However, traditional focus on the mean ADC value (ADC_avg) may oversimplify the complex heterogeneity of breast tumors, and existing studies and meta-analyses often yield contradictory results. For instance, one meta-analysis suggests that lower ADC values are associated with higher Ki-67 expression [8], whereas another meta-analysis indicates that ADC cannot serve as a surrogate marker for Ki-67 expression in breast cancer [9]. Moreover, ADC values may be closely related to tumor cell density as well as the extracellular matrix (ECM) content [10], further complicating the use of ADC in breast cancer diagnosis. Thus, the mean ADC value alone may not adequately capture the diversity and biological characteristics of breast lesions.

In recent years, additional ADC-derived metrics, such as the minimum ADC value (ADC_min), relative minimum ADC ratio (rADC_min), and ADC coefficient of variation (ADC_cv), have been proposed. These metrics provide a more sensitive reflection of the diffusion characteristics within tumors, particularly in regions exhibiting significant heterogeneity. For example, ADC_min is better able to capture areas within tumors that are characterized by high cell density and restricted diffusion [11], while rADC, as a relative value, standardizes the lesion’s ADC against normal tissue, thereby minimizing bias from interindividual tissue variability and improving diagnostic stability and accuracy [12]. Although these advanced ADC metrics demonstrate potential for enhancing the sensitivity and specificity of tumor diagnosis, most existing studies have evaluated these metrics in isolation, lacking comprehensive comparisons and discussions on their combined application.

Therefore, the aim of this study is to retrospectively analyze MRI data of pathologically confirmed breast tumors and systematically assess the diagnostic value of various derived ADC metrics in the differentiation of benign and malignant breast tumors. Furthermore, by constructing a combined predictive model, we seek to further improve the diagnostic accuracy for breast tumors, thereby offering novel approaches for early screening and precision treatment in clinical practice.

Methods

Patients

This retrospective study included pathologically confirmed breast tumor cases from a major medical center in Southwest China, recorded between January 2019 and June 2022. Patients were selected based on the following inclusion criteria: (1) breast tumors diagnosed via preoperative MRI; (2) all lesions on MRI were characterized by mass-like enhancement; (3) definitive pathological diagnosis. The exclusion criteria were as follows: (1) patients who underwent chemotherapy, biopsy, surgery, or other interventions before MRI; (2) incomplete clinical or imaging data; (3) patients with coexisting other malignant tumors; (4) lesions that were too small to delineate and obtain differential results (Fig. 1). Ethical approval was granted by the hospital’s medical ethics committee, and informed consent was waived (Approval No. 2021[5]).

Fig. 1

Flowchart of the patient enrollment process

MRI protocol

MRI scans were performed using a Philips (Amsterdam, Netherlands) Ingenia 1.5 T superconducting scanner equipped with a dedicated breast coil. Patients were positioned prone to allow natural positioning of the breasts within the coil. The scanning protocol included T1-weighted imaging (T1WI), T2-weighted imaging with spectral attenuated inversion recovery (T2WI-SPAIR), DWI, and dynamic contrast-enhanced T1WI (DCE-T1WI). Specific scanning parameters for each sequence are outlined (Table 1). Gadopentetate dimeglumine was used as the contrast agent for DCE-T1WI.

Table 1 MRI scanning parameters

Image processing and data measurement

Routine MR scanning, DWI imaging, and multi-phase dynamic enhancement scanning were performed on all patients. Images were imported into the Philips post-processing workstation for ADC map generation. Utilizing lesion locations from the T2WI-SPAIR and DCE-T1WI, areas with high DWI signal and low ADC signal were marked as the region of interest (ROI), using a b-value of 1000s/mm² in the DWI sequence.

Measurements were conducted jointly by two radiologists with over five years of breast diagnostic experience, resolving any disagreements through discussion. The ROI was manually placed on the largest solid area of the lesion, carefully excluding cystic changes, hemorrhage, necrosis, fat, blood vessels, and artifacts. A circular ROI was used for consistency. The ROI was placed to cover the lesion as evenly as possible, ensuring that the entire lesion area was represented. To avoid bias, the ROI placement was done in a blinded manner, with radiologists unaware of the lesion’s final pathological diagnosis at the time of image assessment. No specific selection of higher or lower ADC values was made during ROI placement. The minimum and average ADC values (ADC_avg) were measured, along with ADC values from the contralateral pectoralis major muscle, normal glandular tissue, and interventricular septum, which were relatively distant from the lesion and minimally influenced by it. The ROI area ranged from 10–50 mm², and measurements were performed on two adjacent slices of the lesion, contralateral pectoralis major muscle, normal glandular tissue, and interventricular septum. For the pectoralis major, the ROI was placed at the center of the muscle to avoid potential signal changes in the peripheral areas. In elderly patients with partial muscle atrophy, the ROI size was proportionally reduced to focus on the preserved muscle fibers. To minimize the impact of cardiac motion artifacts on the measurement of the interventricular septum, the ROI was carefully positioned in the region with the least motion artifact. Additionally, to ensure data accuracy, the physician identified and excluded any outliers during the measurement process, minimizing their impact on the results. An average of three measurements was taken, recording five ADC values for each of the four sites.

The calculation formula was: rADC_min = lesion ADC_min/ADC of the interventricular septum/pectoralis major/glandular tissue (rADC_min_IS, rADC_min_PM, rADC_min_G), obtaining rADC_min values for reference sites and lesion ADC_avg, ADC_min data. Then, the mean (X) and standard deviation (S) of the three ADC measurements were calculated, and the ADC_cv was determined using the formula CV=S/X. These data were subsequently analyzed. The radiological presentations of breast fibroadenoma and invasive ductal carcinoma are depicted in Fig. 2.

Fig. 2

Comparative MR images of breast lesions in two patients. a–d: patient with fibroadenoma. a: T1WI shows lesion with slightly low signal (arrow); b: T2WI-SPAIR shows high signal (arrow); c: DWI shows high signal (arrow); d: ADC map shows slightly low signal, ADC value at 1.64 × 10⁻³ mm²/s (ROI). e–h: patient with invasive ductal carcinoma. e: T1WI shows lesion with low signal (arrow); f: T2WI-SPAIR shows slightly high signal (arrow); g: DWI shows high signal (arrow); h: ADC map shows low signal, ADC value at 0.90 × 10⁻³ mm²/s (ROI)

Statistical analysis

Statistical analyses were conducted using R (version 3.6.3) and Python (version 3.7). Continuous variables that followed a normal distribution were described using mean ± standard deviation, and comparisons between groups were performed using the t-test for equal variances (Levene’s test P ≥ 0.05) and Welch’s t-test for unequal variances (Levene’s test P < 0.05). Non-normally distributed variables were described using median values with the 25th and 75th percentiles (M [P25, P75]) and compared using the Mann-Whitney U test. Subsequently, the diagnostic utility of each of the six ADC metrics was evaluated independently using receiver operating characteristic (ROC) curves, which synchronously calculated sensitivity, specificity, Youden’s index, and the optimal diagnostic threshold for each variable. Correlation analyses among six ADC metrics were performed using Spearman’s method.

Moreover, a combined model incorporating the six ADC metrics and age was developed using a binary logistic regression method with forward and backward stepwise selection based on the Akaike Information Criterion (AIC), and nomogram was generated. The goodness-of-fit, clinical net benefit, and clinical impact of the combined model were assessed using calibration curves, clinical decision curves, and clinical impact curves, respectively. Finally, the comparative analysis of the six ADC metrics and the combined model was performed using the DeLong test, and the improvement in predictive performance for differentiating between benign and malignant breast tumors by the combined model over the optimal single ADC metric was quantified using the integrated discrimination improvement (IDI) index. All statistical tests were two-sided, and P < 0.05 was considered statistically significant.

Results

Pathological results

A total of 125 patients diagnosed with breast tumors were included in this study. The cohort consisted of 77 individuals with malignant breast tumors, predominantly invasive ductal carcinoma, with ages ranging from 28 to 77 years (mean age: 55.66 ± 11.21 years). The remaining 48 patients had benign tumors, mostly fibroadenomas, with ages spanning from 10 to 64 years (mean age: 42.96 ± 14.57 years). In the malignant cohort, 65 cases were identified as non-specific invasive carcinoma, 6 as ductal carcinoma in situ, and 1 case each of intraductal papillary carcinoma, lobular carcinoma, neuroendocrine carcinoma, and adenomyoepithelioma. Additionally, 2 cases of other unspecified malignant tumors were included. The benign cohort comprised 35 cases of fibroadenomas, 6 cases of adenosis, 3 cases of inflammatory changes, 2 intraductal papillomas, 1 case of fibrous hyperplasia, and 1 case of benign phyllodes tumor.

Comparison of six ADC metrics

Statistical analysis revealed significant differences in ADC_avg, ADC_min, ADC_cv and rADC_min values between malignant and benign tumor groups (P < 0.001). Specifically, ADC_min, ADC_avg, and rADC_min values were lower in the malignant group compared to the benign group. Conversely, the ADC_cv value was lower in the benign tumor group (Table 2 and Fig. 3). Spearman correlation analysis revealed significant correlations between ADC_min and ADC_avg, as well as between rADC_min_PM and rADC_min_G, respectively. Conversely, ADC_cv displayed a weak negative correlation with all other metrics (Fig. 4).

Table 2 Comparison of six ADC metrics between benign and malignant breast tumors

Fig. 3

Box plots of six ADC metrics for benign and malignant breast tumors

Fig. 4

Correlation analysis heat map of six ADC metrics

Diagnostic efficacy of single indicators

In the analysis of diagnostic efficacy of the six ADC metrics for differentiating between benign and malignant breast tumors (Table 3 and Fig. 5), ADC_min exhibited the highest area under the curve (AUC) value, along with superior sensitivity and specificity, whereas ADC_cv demonstrated the poorest diagnostic performance. The AUC values of ADC_avg and ADC_min were closely matched, with no statistically significant difference between them as indicated by the DeLong test (Table 4). Therefore, both metrics may be viable options in clinical diagnosis. Furthermore, among the three relative ADC metrics, rADC_min_IS showed the best diagnostic performance, which could be attributed to the greater stability of the interventricular septum density compared to the other two measures.

Table 3 Results of ROC curve analysis of six ADC metrics and combined model

Fig. 5

Roc curves for six ADC metrics and combined model

Table 4 Z and P values for Delong’s test

Diagnostic efficacy of combined model

In the combined model, only age and ADC_avg were included as feature variables for constructing the nomogram (Fig. 6). The model formula is as follows:

$$\log \left( {\frac{p}{{1 - p}}} \right) = 6.061 + 0.126 \times age - 10.832 \times ADC\_avg$$

Fig. 6

Nomogram of the combined model

where p represents the probability of occurrence of malignant breast tumors. The diagnostic performance of the combined model (AUC = 0.964) was superior to that of the individual ADC metrics alone, and the DeLong test revealed statistically significant differences between the diagnostic efficacy of the combined model and all individual ADC metrics. The calibration curve of the combined model showed good agreement with the actual outcomes (Fig. 7), and analyses of clinical decision curves and clinical impact curves also indicated that the nomogram of the combined model achieved a better clinical net benefit (Fig. 8). Finally, the IDI analysis demonstrated that the combined model improved diagnostic performance by 8.3% compared to the optimal single metric (ADC_min), with this result being statistically significant (P = 0.003).

Fig. 7

Calibration curve for combined model

Fig. 8

a: Clinical decision curves for the diagnostic efficacy of the combined model (model1) versus ADC_min (model2) for benign and malignant breast tumors; b: Clinical impact curve of the combined model

Discussion

Breast cancer remains one of the foremost global health challenges. Early and accurate differentiation between benign and malignant lesions is crucial for improving clinical outcomes in breast cancer. DWI and its derived ADC values play a pivotal role in enhancing the diagnostic accuracy of breast MRI. This study contributes to the field by exploring and validating the potential of advanced ADC-derived metrics, including ADC_avg, ADC_min, rADC_min, and ADC_cv, to differentiate benign and malignant breast tumors.

The relationship between ADC values and tumor biology is complex. Numerous studies have highlighted that ADC values are influenced by cellularity, cell membrane integrity, and ECM content [8, 13]. Malignant tumors, with higher cellularity and restricted water diffusion, typically exhibit lower ADC values compared to benign lesions. However, the variability in ADC values across different tumor types and patients underscores the importance of understanding tumor heterogeneity. Despite several meta-analyses on the role of ADC values in assessing breast tumor heterogeneity, contradictions exist between studies. Meyer et al., in their meta-analysis of 28 studies, concluded that ADC values cannot be used to distinguish breast cancer molecular subtypes [14]. In contrast, a meta-analysis by Iima et al., which reviewed 52 studies, indicated that ADC values can serve as molecular biomarkers for estrogen receptor (ER), progesterone receptor (PgR), HER2, and Ki-67 [8]. One reason for this phenomenon may be that previous studies have primarily focused on the diagnostic value of ADC_avg, which may not fully capture tumor heterogeneity [15]. For instance, some breast cancer types, such as mucinous carcinoma, may exhibit higher ADC_avg values despite being malignant, due to their ECM rich in mucin, while certain benign lesions (e.g., abscesses) may show lower ADC_avg values owing to the presence of necrotic tissue, inflammatory cells, and high-protein exudates that restrict water molecule diffusion [7]. In our study, multiple ADC-derived metrics successfully differentiated benign and malignant breast lesions, not only aligning with existing literature, thus reinforcing the validity of ADC as a characteristic biomarker for breast cancer [16,17,18], but also providing a more nuanced approach to interpreting ADC.

A meta-analysis by Surov et al., based on 13,847 lesions, proposed an ADC threshold of 1.0 × 10⁻³ mm²/s as an effective standard for distinguishing benign and malignant breast lesions. This finding is consistent with our study’s ADC_avg threshold (1.04 × 10⁻³ mm²/s) and ADC_min threshold (1.05 × 10⁻³ mm²/s) [7]. Additionally, in our study, ADC_min, as a single metric, demonstrated optimal performance in distinguishing benign from malignant breast tumors, likely due to its sensitive reflection of the densest, most cell-rich, and diffusion-restricted areas of the tumor, providing critical biological information for diagnosis [19, 20]. Moreover, rADC_min, by normalizing ADC values against reference tissue, improved diagnostic consistency across different patients and scanner types, offering a more standardized approach to tumor characterization [13]. Similarly, quantifying the ADC variability within the tumor using ADC_cv may offer insights into tumor heterogeneity, which is essential for assessing tumor invasiveness and predicting treatment response.

Despite these advantages, individual ADC metrics may still face limitations in clinical applications. Therefore, we explored the potential for constructing a combined predictive model. In constructing this model, we chose ADC_avg over ADC_min, likely because ADC_avg provides a more comprehensive assessment of tumor tissue characteristics, reflecting the average cellular density and tissue integrity across the entire tumor area. This approach accounts for tumor heterogeneity and may provide more consistent diagnostic information across different patients and tumor types. Additionally, the selection of ADC_avg over ADC_min in the combined model also considered its overall performance in statistical models [21]. When incorporating age, a critical clinical parameter, ADC_avg likely demonstrated better predictive performance due to its representativeness of the overall tumor characteristics. Through the establishment of the combined model, our study further emphasizes the importance of personalized diagnosis, offering a more accurate tool for breast cancer diagnosis.

While our study explored the potential applications of multiple ADC-derived metrics, standardizing their acquisition, processing, and analysis remains a challenge in clinical practice. The placement of ROI during breast DWI analysis is a complex and bias-prone task. Given the inherent heterogeneity of breast tumors, ROI selection may be influenced by various factors such as tumor location, size, and morphological characteristics [22, 23]. In this study, we used circular ROIs to ensure measurement uniformity and avoid areas affected by necrosis, hemorrhage, or cystic lesions. We also utilized multi-point selection to further improve ADC measurement accuracy. Although this method provided consistent results, the ROI placement strategy must be adjusted based on the specific characteristics of each lesion, due to differences in tumor types and histological features. Thus, standardized scanning protocols and ROI placement methods are crucial to ensuring accuracy and reproducibility of results. Future research should explore ways to standardize ROI placement techniques or introduce automated software to reduce human factors, thereby improving DWI reproducibility and clinical applicability [6].

Furthermore, emerging techniques such as DCE imaging, intravoxel incoherent motion (IVIM), and diffusion kurtosis imaging (DKI) may offer additional possibilities for precise imaging diagnosis of breast cancer. A meta-analysis by Zhang et al. suggested that combining DCE with DWI can further enhance the differentiation of benign and malignant breast lesions [24]. IVIM, by separating water diffusion and microvascular perfusion information, provides a more comprehensive characterization of tumor microstructure [25, 26]. DKI can evaluate the non-Gaussian behavior of water molecule diffusion, further enhancing sensitivity to tumor heterogeneity [27]. Although our study did not include DCE, IVIM, or DKI methods, the combination of routine ADC and its derived metrics with age has already shown high predictive performance. We also acknowledge that DWI combined with these emerging techniques may further improve the diagnostic accuracy and early detection capabilities for breast cancer.

Our study highlights the advantages of integrating DWI into routine breast MRI protocols. Firstly, by offering more precise diagnostic methods, it aids physicians in early-stage differentiation of breast tumor nature, facilitating personalized treatment plans for patients. Secondly, the use of these advanced ADC metrics can reduce unnecessary biopsy procedures, minimizing patient discomfort and treatment costs. Lastly, this study underscores the importance of adopting a multiparametric assessment strategy in breast cancer management, which not only enhances diagnostic accuracy but also provides valuable biomarkers for future therapeutic decisions.

Despite providing valuable insights, this study has two limitations. Its retrospective design and single-center sample may introduce inherent selection biases, limiting the generalizability of the results. Future research should employ prospective, multicenter designs to validate our findings. Additionally, the complexity and heterogeneity of breast cancer were only partially addressed, with a lack of in-depth analysis of tumor grading and molecular subtypes. Future studies in these areas may further guide treatment decisions and improve patient outcomes.

Conclusions

This study highlights the value of advanced ADC metrics in differentiating malignant from benign breast lesions. The combination of ADC metrics and patient age improves diagnostic accuracy and offers potential for enhancing clinical decision-making in breast cancer diagnosis.