Comparison of MR cytometry methods in predicting immunohistochemical factor status and molecular subtypes of breast cancer

Breast cancer stands as one of the most prevalent malignant diseases affecting women, with its incidence and mortality rates rising annually.¹ There are significant differences in terms of malignancy, therapeutic strategies and prognosis across breast cancers with different molecular subtypes.^2–4 Accurate subtype identification is crucial for developing personalized treatment plans for individual patients to reduce mortality and improve prognosis.⁵ Currently, the categorization of breast cancers usually depends on multiple biomarkers from pathological immunohistochemistry (IHC) examinations, and the corresponding subtypes are primarily determined by invasive biopsies with the risks of oedema, bleeding, and infection.⁶ Furthermore, biopsy may lead to missed detection due to the limitation of a localized pathological puncture point, which cannot reflect the overall situation of the lesion.⁷

Imaging techniques that allow for both non-invasive and wide-area detection provide various promising approaches for the diagnosis of breast cancer.⁸ Among them, MRI has been widely used in breast imaging, with its advantages of high image resolution, no ionizing radiation, and multi-contrast imaging, which greatly compensates for the limitations of biopsies and provides more comprehensive information.⁹ Diffusion MRI (dMRI) can reveal tumor microstructures by non-invasively probing the diffusion movement of water molecules without any exogenous contrast agents. The dMRI-derived metric, apparent diffusion coefficient (ADC), has been widely used in the clinical diagnosis¹⁰ and posttreatment evaluation¹¹ of breast cancers. However, this metric represents non-specific, averaged information influenced by several microstructural features with competing effects, reducing diagnostic sensitivity.¹² A meta-analysis¹³ has shown that the ADC values significantly overlapped among different breast cancer subtypes.

Current ADC measurements in clinics usually adopt a single diffusion time t_diff. Over the last decade, there has been an increasing interest in obtaining ADCs with different t_diff. Because the measurement of water diffusion is dependent on t_diff, ADCs with varying probe varying length scales.¹⁴ Therefore, multiple t_diff-based ADC provide more comprehensive information on tissue microstructure. Such a technique is usually termed time-dependent diffusion MRI (t_d-dMRI), which has been widely used in cancer imaging.^15,16 However, time-dependent ADCs still represent the averaged diffusion behavior so it suffers from low specificity of tissue properties. The recently emerging technique of MR cytometry imaging^17–19 provides a potential approach to address the above limitation, which implements multiple-b and multiple-t_diff acquisitions, combined with multi-compartmental biophysical modeling, to decouple different microstructural parameters, such as cell diameter d, intracellular volume fraction v_in, and apparent extracellular diffusivity D_ex. The Imaging Microstructural Parameters Using Limited Spectrally Edited Diffusion (IMPULSED)¹⁷, a specific form of MR cytometry, incorporates both the pulsed-gradient spin-echo (PGSE) and oscillating gradient spin-echo (OGSE)²⁰ acquisitions for a much broader diffusion time range for more comprehensive length scale information, and more importantly, it is readily achievable on 3T MRI scanners with a clinically-feasible scan time of ~6 minutes.¹⁷ For this reason, IMPULSED has been implemented in breast cancer^17,21,22, prostate cancer²³ and gliomas.²⁴

Wang et al.²² have validated the diagnostic efficacy of IMPULSED¹⁷ to classify molecular subtypes of breast cancer. Despite the successful applications, IMPULSED assumes tumor cells as impermeable spheres (without transcytolemmal water exchange), so that dMRI signals from intra-and extracellular compartments are independent of each other, thus greatly simplifying biophysical modeling.^17,25,26 However, neglecting transcytolemmal water exchange will lead to a significant underestimation of intracellular volume fractions and cellularity from MR cytometry.²⁷ On the other hand, the water exchange rate is highly correlated with the cell membrane permeability, which can reflect pathological changes within tumors.^12,28 To obtain this important biophysical information, several methods have been proposed recently, such as JOINT¹⁸ and EXCHANGE.¹⁹ This prospective study is the first to evaluate the efficacy of JOINT and EXCHANGE, which incorporated transcytolemmal water exchange into biophysical modeling, in predicting IHC factor status and molecular subtypes of breast cancer. We compared these methods with time-dependent ADC measurements and IMPULSED.

Determining the IHC factor status and molecular subtypes of breast cancer is an important reference for the development of appropriate clinical treatment regimes. The dMRI-derived ADC metrics have shown potential in the prediction of the IHC factor status and molecular subtypes¹⁰, without the injection of contrast agents in DCE MRI. However, the results of previous studies are controversial. Several publications^10,30,31 reported that the ADC values were lower in ER(+) and PR(+) breast cancers, whereas Park et al.³² showed no significant differences in ADC values between ER(+) and ER(-), as well as PR(+) and PR(-). Besides, some studies^30–33 demonstrated the higher ADC in the HER2(+) group, whereas others have shown the opposite results³⁴ or no significant difference^.35 For Ki67, the conclusion also remains uncertain. Shen et al.³⁶ found that the ADC metrics decreased with higher Ki-67 labeling index, while other findings^33,37 showed that there was no significant difference between high and low Ki67 expression. Some advanced methods, such as intra-voxel incoherent motion (IVIM) imaging^38,39 and diffusionkurtosis imaging (DKI)³⁹, can improve the efficacy to distinguish the status of ER and PR, but failed to differentiate HER2 and Ki67. Recently, Lima et al.⁴⁰ utilized time-dependent ADC measurements obtained by PGSE and OGSE to characterize breast cancer based on IHC markers. Furthermore, Ba et al.²¹ and Wang et al.²² have implemented the emerging MR cytometry method IMPULSED to extract quantitative microstructural information of breast tumors, and the derived parameters has been shown to be effective in the prediction of IHC factor status, molecular subtypes and treatment response to neoadjuvant chemotherapy. However, the biophyscial model used in IMPULSED neglected transcytolemmal water exchange, resulting in the underestimation of v_in and unavailability of membrane permeability.²⁷ This study investigated the efficacy of the MR cytometry methods that incorporate water exchange in predicting IHC factor status and molecular subtypes. The systematic comparisons between the different MR cytometry methods and the t_d-dMRI measurements provide guidance for clinical application of MR cytometry.

In this study, the lower ADC_PGSE and ADC_25Hz values and higher ΔADC were observed in the ER(+) group compared to ER(-), similarly, all three ADC metrics were lower and ΔADC values were higher in the PR(+) group, which are consistent with the previous results.^30,31,40 Some studies^40,41 suggested that the lower ADC values for ER(+) and PR(+) may be due to lower cell membrane permeability, which has been reflected in the results of microstructural parameters, the EXCHANGE-derived transcytolemmal water exchange rate constant k_in was lower in the ER(+) and PR(+) groups, indicating lower membrane permeability. In addition, the IMPULSED and JOINT-derived v_in was larger in the ER(+) group, and all three quantitative methods-derived v_in was larger in the PR(+) group, which may be related to the increased pathological cellularity with ER or PR overexpression in breast tumors.⁴² For the intergroup comparison between HER2(+) and HER2(-), there was no significant difference in ADC-related metrics, this may be due to the fact that HER2 overexpression leads to both increased cell proliferation and angiogenesis, whereas they have opposite impacts on ADC values.³⁰ However, the cell diameter d obtained from all three MR cytometry methods was significantly larger in the HER2(+) group, which was consistent with the pathological finding that HER2-overexpressing breast cancer has increased cell size.⁴³ Finally, for the prediction of Ki67 factor, only cellularity and the EXCHANGE-derived showed significant difference, but we are still unclear about the reasons behind this. On the other hand, for different molecular subtypes, the model-fitted values were the only metrics exhibiting significant difference among these subtypes. Larger d was observed in HER2-enriched subtype compared to non-HER2-enriched subtype, which may be attributed to the HER2 overexpression (larger in the HER(+) group).

In this study, we also systemically compared the diagnosis performance of three MR cytometry methods in predicting IHC factor status and molecular subtypes. For the classifiers based on a single metric, provided the highest AUC in the prediction of PR status and Luminal B subtype; JOINT obtained the highest AUC in predicting HER2 status and HER2-enriched subtype; EXCHANGE performed best in predicting ER, Ki67 status, TNBC and Luminal A subtypes. For the classifiers based on the combined regression model, IMPULSED provided the highest AUC in predicting ER status; JOINT obtained the highest AUC in the prediction of HER2 status and HER2-enriched subtype; EXCHANGE achieved the highest AUC in the prediction of PR status, Ki67 status, TNBC, Luminal A and Luminal B subtypes. The above results show that MR cytometry methods may provide better diagnostic efficacy in the prediction of IHC factor status and molecular subtypes, compared to traditional t_d-dMRI measurements. Meanwhile, the MR cytometry methods incorporating water exchange (JOINT and EXCHANGE) improved the diagnostic efficacy compared to IMPULSED (except for ER status). Although previous numerical simulation and in vitro cell experiments^18,19 demonstrated that JOINT and EXCHANGE, which incorporated water exchange, obtained more accurate estimation of v_in and an additional biophysical parameter k_in, our study found only minor improvements in breast cancer subtyping when water exchange is incorporate into MR cytometry. Thus, while it is desirable to incorporate such objective biophysical phenomena into the biophysical model, improved model accuracy does not necessarily translate into superior clinical diagnostic performance.

There are several limitations in this study. First, the data were collecte in a single center with limited sample size, especially the HER2-enriched subtype. It is necessary to include more breast cancer patients from multiple hospitals or institutions and validate the results more comprehensively. Second, our study lacks the correlation analysis between the MR cytometry-derived parameters and histopathological results. Such analysis will provide more reliable validation on the imaging results and more comprehensive comparisons between the quantitative methods which will be includeD_in our future work. Third, the b values of the OGSE sequence with 50Hz were relatively low (≤500s/mm²) due to the limitations of gradient performance. Despite our best efforts to eliminate the impact of IVIM before model fitting, molecular markers of angiogenesis such as micro-vessel density⁴⁴ still introduce bias in the estimation of microstructural parameters, especially when using low b values. Fortunately, the modern whole-body ultra-high-performance gradients can provide higher b values for high-frequency OGSE when PNS allows. Fourth, each imaging metric was averaged across the whole ROI, which lost the information of spatial heterogeneity within breast tumors. Surviving cells, dead cells and necrotic regions may co-exist in each ROI. This spatial heterogeneity can be captured by histogram analysis.³⁸

In summary, this study was the first to evaluate the clinical performance of MR cytometry incorporating water exchange in predicting IHC factor status and molecular subtypes of breast cancer, and comprehensively compared the derived microstructural parameters obtained and conventional t_d-dMRI metrics. Our results showed that advanced MR cytometry outperformed traditional ADC measurements. Incorporating water exchange into MR cytometry methods further improved the diagnosis performance. Specifically, the results based on the multi-variable regression models showed that: IMPULSED performed best in predicting ER status; JOINT was more suitable for predicting HER2 status and HER2-enriched subtype; EXCHANGE can provide the highest AUC in predicting PR and Ki67 status, TNBC, Luminal A and Luminal B subtypes.