Comparison of MR cytometry methods in predicting immunohistochemical factor status and molecular subtypes of breast cancer
Categoría del artículo: Research Article
Publicado en línea: 06 ago 2025
Páginas: 337 - 348
Recibido: 03 mar 2025
Aceptado: 19 may 2025
DOI: https://doi.org/10.2478/raon-2025-0044
Palabras clave
© 2025 Lei Wu et al., published by Sciendo
This work is licensed under the Creative Commons Attribution 4.0 International License.
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
Wang
This study was approved by the institution review board of the local hospital. A total of 96 patients with breast cancer who underwent MRI from May 2023 to April 2024 were prospectively collected, with the inclusion criteria as follows: (1) unilateral lesions with a diameter >1 cm; (2) no prior treatment for any breast disease before the MRI examination, including biopsy, neoadjuvant therapy, and other anti-tumor treatments; (3) complete clinical pathological data available after the MRI examination. Six of ninety-six cases were excluded due to failure in fat suppression (n = 3), motion artifacts (n = 2) and multi-focal lesions (n = 1).
Breast dMRI acquisitions were conducted on a 3T MR scanner with a maximum gradient amplitude of 80 mT/m and a maximum slew rate of 200 T/m/s (MAGNETOM Prisma, Siemens Healthineers, Forchheim, Germany). A dedicated 16-channel phased-array breast coil was employed. A combined acquisition protocol lasting 6 minutes was used to obtain diffusion images with different diffusion times, which included the PGSE (effective diffusion time
Immunohistochemistry and fluorescence in situ hybridization (FISH) were performed on each participant followed by MRI examination. According to the obtained status (positive or negative) of Estrogen Receptor (ER), Progesterone Receptor (PR), Human Epidermal Growth Factor Receptor-2 (HER2) and proliferation marker (Ki67), the lesions were classified into Luminal A, Luminal B, HER2-enriched and Triple-Negative Breast Cancer (TNBC) subtypes (Supplementary Appendix 2).
Three MR cytometry methods were used to fit microstructural parameters from the acquired dMRI signals: IMPULSED, JOINT and EXCHANGE, please refer to Supplementary Materials for more details. There are four free parameters in IMPULSED, including intracellular volume fraction
For the conventional td-dMRI measurements, ADC values at each sequence were obtained by fitting the multi-b signals to
Regions of interest (ROIs) were manually drawn on the diffusion-weighted images by radiologists L. W. and H. B. (with six years of experience), who were blinded to the final pathological results. Surrounding fatty, muscle, cystic, hemorrhagic and necrotic tissues were carefully excluded from the ROIs. The time-dependent ADC metrics and model-fitted microstructural parameters were calculated within each voxel and then averaged over the whole ROI.
Statistical analysis methods varied by grouping strategies.
There are four binary classification tasks,
This is a four-class classification task,
A total of 90 patients with 90 lesions were included in this prospective study. Among these, 26 cases were Luminal A subtype, 38 cases were Luminal B subtype, 8 cases were HER2-enriched subtype, and 18 cases were TNBC subtype. The patient information and lesion characteristics are summarized in Table 1.
Patient information and lesion characteristics
Characteristics | Luminal A (n = 26) | Luminal B (n = 38) | TNBC (n = 18) | HER2-enriched (n = 8) |
---|---|---|---|---|
Age(years) | 55.11 ± 8.52 | 51.16 ± 11.01 | 52.89 ± 8.45 | 51.00 ± 9.04 |
Tumor size(mm) | 27.65 ± 6.93 | 27.37 ± 8.25 | 27.83 ± 9.93 | 24.50 ± 6.35 |
Menstruation state | ||||
Premenopausal women | 11 | 17 | 5 | 3 |
Postmenopausal women | 15 | 21 | 13 | 5 |
Tumor border | ||||
Well-defined | 9 | 10 | 8 | 4 |
ill-defined | 17 | 28 | 10 | 4 |
Tumor sharp | ||||
Oval or round | 21 | 32 | 11 | 3 |
Irregular | 5 | 6 | 7 | 5 |
ER status | ||||
Positive | 26 | 38 | 0 | 0 |
Negative | 0 | 0 | 18 | 8 |
PR status | ||||
Positive | 24 | 29 | 0 | 0 |
Negative | 2 | 9 | 18 | 8 |
HER2 status | ||||
Positive | 0 | 13 | 0 | 8 |
Negative | 26 | 25 | 18 | 0 |
Ki67 status | ||||
Positive | 3 | 27 | 16 | 6 |
Negative | 23 | 11 | 2 | 2 |
ER = estrogen receptor; HER2 = human epidermal growth factor receptor 2; Ki67 = nuclear associated antigen; PR = progesterone receptor; TNBC = triple-negative breast cancer
Supplementary Figure 1 shows the PGSE and OGSE diffusion-weighted images for a representative patient (Luminal B subtype). The lesion conspicuity is acceptable. Figure 1 shows the ADC metrics and microstructural parameter maps for five representative breast cancer patients with different IHC factor status and molecular subtypes. The overall ADC values decreased with longer diffusion time,

The ADC and microstructural maps overlaid on b = 1000 s/mm2 diffusion-weighted images of five representative breast cancer patients.
ADC = apparent diffusion coefficient; ER = estrogen receptor;
Figure 2 shows the results of intergroup comparisons in the IHC factors. For td-dMRI measurements, ADCPGSE and ADC25Hz were significantly lower in ER(+) cases, and was higher (

Intergroup comparison of td-MRI metrics and microstructural parameters respectively fitted from IMPULSED, JOINT and EXCHANGE between positive and negative immunohistochemical factor status.
* =
Table 2 shows the results of intergroup comparisons across the four molecular subtypes. The cell diameter was the only metric that showed significant difference across the four breast cancer molecular subtypes (
The intergroup comparison for the imaging metrics across four breast cancer molecular subtypes
Model | Parameter | TNBC Median (IQR) | HER2-enriched Median (IQR) | Luminal A Median (IQR) | Luminal B Median (IQR) | |
---|---|---|---|---|---|---|
td-dMRI | ADCPGSE | 0.85 (0.49) | 0.90 (0.25) | 0.74 (0.25) | 0.81 (0.25) | 0.106 |
ADC25Hz | 1.10 (0.51) | 1.21 (0.31) | 1.03 (0.19) | 1.08 (0.22) | 0.055 | |
ADC50Hz | 1.44 (0.52) | 1.53 (0.26) | 1.38 (0.18) | 1.38 (0.18) | 0.071 | |
ΔADC | 0.73 (0.38) | 0.66 (0.29) | 0.81 (0.41) | 0.83 (0.42) | 0.075 | |
IMPULSE | 15.00 (1.73) | 16.16 (1.85) | 14.79 (1.33) | 14.96 (1.07) | ||
Vin | 0.38 (0.13) | 0.37 (0.07) | 0.42 (0.13) | 0.41 (0.09) | 0.063 | |
Dex | 1.91 (0.54) | 2.08 (0.28) | 2.02 (0.24) | 1.95 (0.32) | 0.712 | |
Din | 2.09 (0.20) | 2.15 (0.08) | 2.07 (0.31) | 2.05 (0.21) | 0.598 | |
Cellularity | 0.074 (0.03) | 0.058 (0.03) | 0.078 (0.05) | 0.075 (0.03) | 0.071 | |
JOIN | d | 15.73 (1.57) | 17.17 (2.09) | 15.65 (2.02) | 16.05 (1.42) | |
vin | 0.51 (0.15) | 0.51 (0.09) | 0.54 (0.09) | 0.55 (0.07) | 0.144 | |
kin | 18.12 (6.66) | 16.38 (2.89) | 15.68 (6.75) | 16.74 (5.09) | 0.374 | |
Dex | 2.35 (0.41) | 2.54 (0.22) | 2.40 (0.26) | 2.37 (0.32) | 0.596 | |
Cellularity | 0.081 (0.04) | 0.063 (0.03) | 0.083 (0.06) | 0.082 (0.03) | 0.114 | |
EXCHANGE | d | 13.91 (1.31) | 15.07 (1.79) | 13.67 (1.20) | 13.94 (1.07) | |
vin | 0.58 (0.10) | 0.58 (0.07) | 0.58 (0.05) | 0.59 (0.05) | 0.280 | |
kin | 8.12 (5.50) | 7.00 (3.44) | 6.70 (3.9) | 6.67 (2.2) | 0.061 | |
Dex | 2.30 (0.45) | 2.52 (0.20) | 2.36 (0.31) | 2.30 (0.34) | 0.442 | |
Cellularity | 0.12 (0.03) | 0.09 (0.03) | 0.12 (0.08) | 0.12 (0.03) | 0.053 |
The numbers in bold represent there is a significant difference across four molecular subtypes.
ADC = apparent diffusion coefficient; ER = estrogen receptor;
Table 3 shows the AUC values for the prediction of IHC factor status. For ER, EXCHANGE-derived

The performance of derived parameters in predicting immunohistochemistry (IHC) factor status. In each sub-plot, the four curves respectively correspond to: the classifier with the highest AUC based on a single td-dMRI metric (ADCPSGE, ADC25Hz, ADC50Hzor ΔADC), the classifier based on the combination of all td-dMRI metrics, the classifier with the highest AUC based on a single model-fitted microstructural parameter (
ADC = apparent diffusion coefficient; ER = estrogen receptor;
The diagnostic performance of imaging metrics for the prediction of immunohistochemistry (IHC) factor status
Model | Parameter | AUC (ER) | AUC (PR) | AUC (HER2) | AUC (Ki67) |
---|---|---|---|---|---|
td-dMRI | ADCPGSE | 0.631 (0.508, 0.755) | 0.693 (0.584, 0.803) | 0.594 (0.470, 0.718) | 0.553 (0.427, 0.678) |
ADC25Hz | 0.630 (0.508, 0.752) | 0.682 (0.571, 0.793) | 0.639 (0.055, 0.767) | 0.580 (0.458, 0.702) | |
ADC50Hz | 0.624 (0.493, 0.755) | 0.674 (0.560, 0.788) | 0.627 (0.500, 0.755) | 0.571 (0.449, 0.693) | |
ΔADC | 0.660 (0.540, 0.779) | 0.468 (0.328, 0.608) | 0.496 (0.369, 0.623) | ||
Combined | 0.645 (0.522, 0.768) | 0.688 (0.576, 0.800) | 0.623 (0.476, 0.770) | 0.633 (0.516, 0.750) | |
IMPULSED | 0.590 (0.454, 0.726) | 0.621 (0.501, 0.742) | 0.652 (0.512, 0.793) | 0.612 (0.494, 0.730) | |
Vin | 0.664 (0.550, 0.779) | 0.686 (0.576, 0.796) | 0.554 (0.433, 0.675) | 0.545 (0.419, 0.670) | |
Dex | 0.529 (0.389, 0.669) | 0.587 (0.461, 0.714) | 0.518 (0.337, 0.659) | 0.558 (0.438, 0.679) | |
Din | 0.540 (0.407, 0.673) | 0.595 (0.473, 0.716) | 0.567 (0.433, 0.700) | 0.524 (0.399, 0.649) | |
Cellularity | 0.646 (0.521, 0.771) | 0.638 (0.519, 0.758) | 0.567 (0.426, 0.708) | 0.638 (0.521, 0.754) | |
Combined | 0.705 (0.597, 0.813) | 0.689 (0.552, 0.826) | 0.646 (0.532, 0.760) | ||
JOIN | d | 0.575 (0.443, 0.707) | 0.601 (0.481, 0.721) | 0.595 (0.476, 0.714) | |
vin | 0.643 (0.523, 0.764) | 0.673 (0.559, 0.787) | 0.453 (0.330, 0.577) | 0.517 (0.394, 0.641) | |
kin | 0.623 (0.507, 0.740) | 0.535 (0.415, 0.655) | 0.459 (0.335, 0.583) | 0.520 (0.392, 0.649) | |
Dex | 0.487 (0.351, 0.623) | 0.601 (0.478, 0.724) | 0.536 (0.399, 0.673) | 0.524 (0.403, 0.646) | |
Cellularity | 0.619 (0.490, 0.747) | 0.613 (0.491, 0.736) | 0.577 (0.438, 0.716) | 0.632 (0.513, 0.750) | |
Combined | 0.731 (0.625, 0.837) | 0.718 (0.609, 0.827) | 0.666 (0.552, 0.781) | ||
EXCHANGE | d | 0.584 (0.450, 0.718) | 0.624 (0.504, 0.744) | 0.650 (0.510, 0.790) | |
vin | 0.596 (0.466, 0.725) | 0.671 (0.555, 0.788) | 0.511 (0.380, 0.642) | 0.466 (0.343, 0.590) | |
kin | 0.643 (0.526, 0.760) | 0.528 (0.407, 0.650) | 0.547 (0.420, 0.675) | ||
Dex | 0.521 (0.382, 0.661) | 0.608 (0.483, 0.732) | 0.562 (0.424, 0.699) | 0.522 (0.401, 0.643) | |
Cellularity | 0.618 (0.490, 0.745) | 0.617 (0.496, 0.739) | 0.594 (0.445, 0.732) | 0.632 (0.515, 0.748) | |
Combined | 0.725 (0.610, 0.839) | 0.668 (0.542, 0.794) |
AUC values are presented as mean (bootstrapped 95% CIs). The numbers in bold represent the highest AUC values respectively achieved by the single-variable regression model and multi-variable (combined) regression model. In the combined model, all the parameters obtained by each method was included.
ADC = apparent diffusion coefficient;
Table 4 shows the AUC values for the prediction of molecular subtypes of breast cancers. For TNBC, EXCHANGE-derived

The performance of derived parameters in predicting breast cancer molecular subtypes. In each sub-plot, the four curves respectively correspond to: the classifier with the highest AUC based on a single td-dMRI metric (ADCPSGE, ADC25Hz, ADC50Hz or ΔADC), the classifier based on the combination of all td-dMRI metrics, the classifier with the highest AUC based on a single model-fitted microstructural parameter (
ADC = apparent diffusion coefficient; AUC = area under the receiver operating characteristic curve; TNBC = triple-negative breast cancer;
The diagnostic performance of imaging metrics for the prediction of molecular subtypes
Model | Parameter | AUC (TNBC) | AUC (HER2- enriched) | AUC (Luminal A) | AUC (Luminal B) |
---|---|---|---|---|---|
ADC | ADCPGSE | 0.617 (0.470, 0.763) | 0.681 (0.519, 0.844) | 0.570 (0.438, 0.703) | 0.577 (0.458, 0.697) |
ADC25Hz | 0.518 (0.435, 0.727) | 0.745 (0.614, 0.877) | 0.600 (0.470, 0.729) | 0.551 (0.429, 0.672) | |
ADC50Hz | 0.575 (0.411, 0.739) | 0.744 (0.624, 0.863) | 0.576 (0.449, 0.703) | 0.566 (0.446, 0.686) | |
ΔADC | 0.648 (0.511, 0.785) | 0.360 (0.141, 0.579) | 0.474 (0.340, 0.609) | ||
Combined | 0.644 (0.501, 0.786) | 0.765 (0.623, 0.907) | 0.659 (0.538, 0.781) | 0.633 (0.517, 0.748) | |
IMPULSED | 0.519 (0.316, 0.676) | 0.784 (0.609, 0.958) | 0.614 (0.487, 0.741) | 0.490 (0.370, 0.610) | |
Vin | 0.657 (0.522, 0.793) | 0.651 (0.489, 0.813) | 0.572 (0.433, 0.711) | 0.593 (0.475, 0.710) | |
Dex | 0.537 (0.367, 0.707) | 0.582 (0.375, 0.790) | 0.565 (0.445, 0.684) | 0.558 (0.436, 0.680) | |
Din | 0.507 (0.348, 0.666) | 0.622 (0.468, 0.776) | 0.514 (0.376, 0.653) | 0.533 (0.412, 0.654) | |
Cellularity | 0.593 (0.447, 0.738) | 0.720 (0.503, 0.936) | 0.606 (0.474, 0.737) | 0.455 (0.336, 0.574) | |
Combined | 0.748 (0.629, 0.868) | 0.739 (0.531, 0.947) | 0.666 (0.544, 0.789) | 0.630 (0.513, 0.747) | |
JOIN | d | 0.519 (0.367, 0.671) | 0.590 (0.460, 0.719) | 0.515 (0.394, 0.635) | |
vin | 0.644 (0.496, 0.791) | 0.611 (0.438, 0.785) | 0.545 (0.412, 0.678) | 0.593 (0.475, 0.712) | |
kin | 0.630 (0.489, 0.772) | 0.486 (0.349, 0.624) | 0.558 (0.414, 0.701) | 0.541 (0.420, 0.663) | |
Dex | 0.521 (0.363, 0.679) | 0.642 (0.438, 0.845) | 0.507 (0.383, 0.631) | 0.539 (0.417, 0.662) | |
Cellularity | 0.549 (0.396, 0.703) | 0.733 (0.537, 0.929) | 0.584 (0.450, 0.718) | 0.461 (0.342, 0.580) | |
Combined | 0.742 (0.616, 0.869) | 0.648 (0.525, 0.770) | 0.609 (0.492, 0.727) | ||
EXCHANGE | d | 0.509 (0.357, 0.661) | 0.784 (0.602, 0.965) | 0.516 (0.396, 0.636) | |
vin | 0.627 (0.477, 0.778) | 0.532 (0.309, 0.755) | 0.492 (0.364, 0.621) | 0.601 (0.481, 0.721) | |
kin | 0.459 (0.299, 0.618) | 0.543 (0.402, 0.684) | 0.606 (0.489, 0.723) | ||
Dex | 0.478 (0.313,0.644) | 0.666 (0.468, 0.865) | 0.514 (0.390, 0.637) | 0.553 (0.431, 0.674) | |
Cellularity | 0.542 (0.393, 0.692) | 0.756 (0.559, 0.953) | 0.620 (0.490, 0.750) | 0.488 (0.368, 0.608) | |
Combined | 0.784 (0.598, 0.969) |
AUC values are presented as mean (bootstrapped 95% CIs). The numbers in bold represent the highest AUC values respectively achieved by the single-variable regression model and multi-variable (combined) regression model. In the combined model, all the parameters obtained by each method was included.
ADC = apparent diffusion coefficient;
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 publications10,30,31 reported that the ADC values were lower in ER(+) and PR(+) breast cancers, whereas Park
In this study, the lower ADCPGSE and ADC25Hz 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 studies40,41 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
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 td-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 experiments18,19 demonstrated that JOINT and EXCHANGE, which incorporated water exchange, obtained more accurate estimation of
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 includeDin 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 td-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.