Skip to main content
Download PDF

Quantification of intratumoral heterogeneity using habitat-based MRI radiomics to identify HER2-positive, -low and -zero breast cancers: a multicenter study

Breast Cancer Research volume 26, Article number: 160 (2024) Cite this article

A Correction to this article was published on 29 November 2024

This article has been updated

Abstract

Background

Human epidermal growth factor receptor 2-targeted (HER2) therapy with antibody-drug conjugates has proven effective for patients with HER2-low breast cancer. However, intratumoral heterogeneity (ITH) poses a great challenge in identifying HER2-low tumors. ITH signatures were developed by quantifying ITH to differentiate HER2-positive, -low and -zero breast cancers.

Methods

This retrospective study included 614 patients from two institutions. The study was structured into two primary tasks: task 1 was to differentiate between HER2-positive and -negative tumors, followed by task 2 to differentiate HER2-low and -zero tumors. Whole-tumor radiomics features and habitat radiomics features were extracted from MRI to construct the radiomics and ITH signatures. Multivariable logistic regression analysis was used to determine significant independent predictors. A combined model integrating significant clinicopathologic variables, radiomics signature, and ITH signature was developed for task (1) Subsequently, the better-performing model was established using the same approach for task (2) The area under the receiver operating characteristic curve (AUC) was used to assess the performance of each model.

Results

Task 1 comprised 614 patients (training, n = 348; validation, n = 149; and test cohorts, n = 117). Task 2 encompassed 501 patients (training, n = 283; validation, n = 122; and test cohorts, n = 96). For task1, the ITH signature showed outstanding performance, achieving AUCs of 0.81, 0.81, and 0.81 in the training, validation and test cohorts, respectively. The combined model achieved improved performance, with AUCs of 0.83, 0.84 and 0.83 across the three cohorts, respectively. For task2, the ITH signature maintained superior performance, with AUCs of 0.94, 0.93 and 0.84 across the training, validation and test cohorts, respectively. Multivariable logistic regression analysis indicated that none of the clinicopathologic characteristics were retained as predictors associated with odds of HER2-low tumors.

Conclusions

Our study developed ITH signatures that quantified ITH using habitat-based MRI radiomics, achieving outstanding performance in differentiating HER2-postive and -negative tumors, and further differentiating HER2-low and -zero breast cancers.

Introduction

Human epidermal growth factor receptor 2 (HER2) stands as a quintessential oncogene characterized by its robust cell proliferation, rendering it a prime therapeutic target for breast cancers marked by HER2 gene amplification or HER2 protein overexpression [1]. Despite the benefits of current therapeutic interventions, these have not yet translated to patients with HER2-negative tumors [2]. HER2-negative expression can be further categorized into HER2-zero and HER2-low based on the absent or low expression of HER2. It has been demonstrated that HER2-low tumors exhibit more clinically malignant and invasive characteristics compared to their HER2-zero counterparts [3, 4]. However, recent studies have unveiled a prospective improvement in prognosis for patients with metastatic HER2-low breast cancer, particularly through the utilization of trastuzumab deruxtecan, a pioneering antibody-drug conjugate tailored to target HER2 [5,6,7]. This emerging body of evidence holds promise for altering the treatment pattern for patients with HER2-low breast cancer.

The current HER2 testing algorithm, based on immunohistochemistry (IHC) and/or fluorescence in situ hybridization (FISH), can detect the HER2-low subtype (IHC score 1+, or IHC 2 + without FISH amplification) [8]. Nevertheless, this standardized pathological technique has been reported to have low reproducibility, with variations observed in the accuracy of -zero and -low status [9]. This inconsistency is attributed to intratumoral heterogeneity (ITH); thus, single-site biopsies are inadequate for comprehensively characterizing the genetic and pathological characteristics. Moreover, ITH accounts for distinctions in the response of breast tumors to identical treatment strategies, particularly among diagnostically similar patients [10]. Significant ITH of HER2 expression has been revealed in HER2-low breast cancer [11]. Hence, it is imperative to employ a method capable of characterizing tumor heterogeneity to identify HER2-low tumors that may exhibit a potential response to treatment.

Quantitative analysis of ITH based on breast MRI has demonstrated considerable utility in assessing treatment response to neoadjuvant therapy and prognosis [12, 13]. While several recent studies have shown an association between radiomics and HER2-low breast cancer [14,15,16], features selected based on the entire tumor or peritumor were still unable to reflect ITH. In contrast to classic radiomics, ITH was captured by extracting radiomics features from intratumoral habitats. Habitats can be conceptualized as the division of tumors into subregions, which reflects more diverse voxel intensity information to depict ITH [17]. Based on the considerations mentioned, we hypothesize that this quantitative technique for measuring ITH can accurately identify HER2-low tumors.

Therefore, we aim to construct models based on a quantitative characterization of ITH to differentiate between HER2-positive and -negative tumors, and to further distinguish HER2-low and -zero tumors.

Methods

Patients

This retrospective multicenter study was approved by the institutional review board and ethics committee for review of Peking University People’s Hospital and Jiangmen Central Hospital, and the requirement for informed consent was waived in both institutions.

A total of 594 consecutive patients from Center 1 (Peking University People’s Hospital) between July 2017 and April 2019, and 159 consecutive patients from Center 2(Jiangmen Central Hospital)between July 2019 and October 2021 were included. The inclusion criteria comprised the following: [1] female patients diagnosed with primary invasive breast cancer; [2] patients who underwent dynamic contrast-enhanced MRI (DCE-MRI); and [3] HER2 status was determined by the IHC and/or FISH of postsurgical specimens. The exclusion criteria were as follows: [1] previous neoadjuvant therapy, radiotherapy, or other treatments before MRI scan; [2] incomplete or unavailable HER2 assessment results; [3] incomplete or unavailable clinical data; and [4] poor quality of MR images, including significant DWI deformation, insufficient image resolution and significant artifacts, which makes it difficult for experienced radiologists to diagnose a lesion or no obvious lesions observed.

Finally, a total of 614 patients were enrolled. The study aimed to identify patients with HER2-low breast cancer and involved two tasks. Task 1 focused on distinguishing between HER2-positive and -negative status. From center 1, 497 patients were non-selectively divided into two cohorts at a ratio of 7:3 for each model (training cohort, n = 348; validation cohort, n = 149). Task 2 was performed to differentiate between HER2-low and -zero status. Four hundred five patients with HER2-negative expression from Center 1 were enrolled for model construction (training cohort, n = 283; validation cohort, n = 122) using the same ratio. Additionally, 117 patients from Center 2 were enrolled for external testing. All these patients were included in task 1, and 96 patients with HER2-negative expression were included in task 2. The patient enrollment pathway is illustrated in Fig. 1.

Fig. 1
figure 1

Flowchart of patient inclusion and exclusion criteria in the study. HER2, human epidermal growth factor receptor 2

Full size image

Clinicopathologic data

Both centers obtained clinical data from the institution archives, including gender, age, menopausal status, clinical T stage, and N stage. HER2 status was detected through IHC according to the 2018 American Society of Clinical Oncology and College of American Pathologists testing guidelines [8]. Further FISH testing was performed when HER2 status was determined to be 2 + results. HER2 status were grouped as follows: HER2-zero (IHC score of 0), HER2-low (IHC score of 1 + or IHC 2 + without FISH amplification), and HER2-positive (IHC score of 2 + with FISH amplification or 3+). The details of other histopathologic dada, including histologic type, histology grade, Ki-67 expression, and hormone receptor (HR) status, are described in Additional File 1.

MRI segmentation and habitat generation

Details regarding the MRI protocol and parameters at the two institutions are provided in the Additional File 1. The diffusion-weighted imaging (DWI) was initially acquired prior to contrast administration, followed by a DCE-MRI consisting of an axial T1-weighted fat-suppressed sequence. All images were exported from the picture archiving and communication systems (PACS) and used for analysis, followed by resampling with a voxel spacing of 1 × 1 × 1 mm3, and histogram normalization of intensity values were applied to correct differences among MR scanners.

Segmentation was performed using DWI (b value of 800 s/mm2) and the first phase contrast-enhanced T1-weighted imaging (T1C) after the injection. The delineation of the volume of interest (VOI) on images was meticulously executed using ITK-SNAP software (version 3.8.0; http://www.itksnap.org/pmwiki/pmwiki.php) by two experienced radiologists together (F.C. and X.X.J.; with 5 and 8 years of breast MRI experience, respectively), both of whom were blinded to clinicopathologic data. In cases where the radiologists disagreed, a senior radiologist (Y.W. 20 years of breast MRI experience) reviewed the images and provided the final delineation. The VOIs were carefully delineated layer by layer along the boundaries of the tumor profile to encompass the entire tumor volume.

For tumor habitat generation, the simple linear iterative clustering (SLIC) [18] was employed to divide each tumor into 100 super-pixels. Nineteen radiomics features extracted from each super-pixel are detailed in the Additional File 1. Subsequently, the K-means method was applied to further cluster habitats in this study (Fig. 2). A range of cluster counts from 3 to 10 were evaluated. The optimal number was selected based on the Calinski-Harabasz (CH) score [19]. Additional details of our methodology for habitat generation can be found in Additional File 1. Our clustering analysis was conducted using the scikit-learn package in Python (version 3.0.1; https://pyradiomics.readthedocs.io).

Fig. 2
figure 2

Workflow of this study. (I) VOI segmentation: Segmentation in DWI and T1C. (II) Habitats generation: SLIC to generate 100 super-pixels, and K-means methods for clustering habitats. (III) Features extraction: K nearest neighbor approach to address missing data. Radiomics features including shape, histogram and texture features. (IV) Features selection: Statistical evaluation, Pearson’s correlation and LASSO analysis for selecting the most important features. (V) Model evaluation: AUC and 95%CI for evaluating the performance, DCA for assessing the clinical net benefit, and calibration curve to evaluate the goodness of fit of the optimal model. (VI) Model interpretation was given by SHAP. VOI, volume of interest, DWI, diffusion-weighted imaging, T1C, contrast-enhanced T1-weighted sequence, SLIC, Simple Linear Iterative Clustering, LASSO, least absolute shrinkage and selection operator, AUC, area under the receiver operating characteristic curve, DCA, decision curve analysis, SHAP, shapley additive explanations analysis

Full size image

Radiomics feature extraction and selection

1834 radiomics features were extracted independently from specified habitats and whole-tumor region within DWI and T1C and were subsequently employed for further selection. The details of habitat radiomics features and whole-tumor radiomics features are described in Additional File 1. For areas with insufficient voxel counts, a k-nearest neighbor approach was employed to address missing data, ensuring the robustness and reliability of the analysis. All features were extracted using Pyradiomics version 3.0.1 (http://pyradiomics.readthedocs.io) [20], following the guidelines of the Imaging Biomarker Standardization Initiative [21].

The distribution of features was normalized using the mean and standard deviation of the training cohort. Statistical evaluation of features was performed using t-tests, with a significance threshold set at P < 0.05 to retain only statistically significant features. Highly correlated features were eliminated using Pearson’s correlation, with the threshold set at 0.9. The least absolute shrinkage and selection operator (LASSO) regression was then applied to perform multifactor analysis for features dimension reduction. The optimal regularization parameter λ was determined through 10-fold cross-validation to identify the most predictive features.

Model development and interpretation

For task 1, univariable logistic regression was used to evaluate clinicopathologic variables associated with HER2 status, and variables with P < 0.05 were fed into the multivariable logistic regression with the Akaike information criterion. Clinicopathologic variables identified as independent predictors after multivariable logistic regression were utilized to develop a clinical model. The selected whole-tumor radiomics features were used to construct the radiomics signature, while habitat radiomics features were employed to established the ITH signature. Additionally, a combined model integrating significant clinicopathologic variables, the radiomics signature, and the ITH signature were developed. The superior model was further established for task 2. The model development process was supervised, and five classification algorithms supported by the Onekey AI platform (version 3.1.8) were employed, including Logistic Regression (LR), Random Forest (RF), Extratrees, XGBoost, and LightGBM. Model interpretation was facilitated through shapley additive explanations analysis (SHAP) [22] to visualize the impact of habitat radiomics features on the prediction of the models.

Statistical analysis

Categorical variables were expressed as frequencies with percentages and analyzed using the χ2 test or Fisher exact test. The performance of the models was assessed by the area under the curve of the receiver operator characteristic (AUC) and its corresponding 95% confidence interval (CI). Optimal thresholds for identifying HER2 expressions were derived using the Youden index. Sensitivity, specificity, accuracy and F1 score were also calculated to detail the performance of models. Decision curves analysis (DCA) was used to evaluate the clinical benefit of the predictive models. The calibration curve was used to evaluate the goodness of fit of the optimal model. All statistical tests were two-sided, with a significance level of P < 0.05. Statistical analysis was performed using R (version 3.4.3, https://www.r-project.org), Python (version 3.7.1, https://www.python.org/) and SPSS (version 26.0, https://www.ibm.com/spss).

Results

Patient characteristics

A total of 614 female patients were included in the study from two institutions (mean age 54 ± 12). The prevalence rates of HER2-positive and HER2-low status were 18.40% (113/614) and 50.33% (309/614), respectively. For task 1, differences in histomolecular class and HR status were observed across three cohorts. The HER2-positive tumors had higher Ki-67 expression than the HER2-negative tumors in the training and validation cohorts. The HER2-positive tumors had higher clinical T stage than the HER2-negative tumors only in the test cohort and had a higher grade in the validation and test cohorts. None of the remaining clinicopathologic characteristics were statistically different (Table 1). For task 2, differences in HR status and histomolecular class were observed in the training and test cohort. The HER2-low tumors had lower Ki-67 expression than the HER2-zero tumors in the training and validation cohorts. None of the remaining clinicopathologic characteristics were different (Table 2).

Table 1 Clinicopathologic characteristics of study patients in task 1
Full size table
Table 2 Clinicopathologic characteristics of study patients in task 2
Full size table

Feature extraction and selection

The generation of habitat zones was visualized in Fig. 3. The VOI of the tumor was segmented into 100 super pixels, which were subsequently clustered to delineate intratumoral habitat zones. The number 3 emerged as the best cluster for habitat analysis according to the results of CH score (Figure S1, Additional File 1). Features were extracted from three intratumoral habitats within DWI and T1C, resulting in a total of 11,004 habitat radiomics features for further selection (Figure S2, Additional File 1). Similarly, 3,668 whole-tumor radiomics features were extracted from two sequences.

Fig. 3
figure 3

Example super pixels segmentation and habitats generation of patients with HER2-zero, -low and -positive breast cancers. (a) A 42-year-old woman with HER2-zero breast cancer; (b) A 49-year-old woman with HER2-low breast cancer; and (c) A 56-year-old woman with HER2-positive breast cancer. Tumors reflect different super pixels and habitats by different colors. The habitat zones were generated separately in T1C and DWI. Habitat zones generated in both tasks from the same patient were largely consistent, demonstrating the stability of habitats. HER2, human epidermal growth factor receptor 2, T1C, contrast-enhanced T1-weighted imaging, DWI, diffusion-weighted imaging, VOI, volume of interest

Full size image

After feature dimension reduction by Lasso (Figure S3, Additional File 1), for task 1, sixteen most important whole-tumor radiomics features were chosen to develop the radiomics signature, while thirteen most important habitat radiomics features were selected for the ITH signature (Table S2, Additional File 1). For task 2, nineteen most important habitat radiomics features were selected for constructing the ITH signature (Table S3, Additional File 1).

Variables associated with odds of HER2-positive and HER2-low status

For task 1, univariable analysis of variables unveiled a significant association between several variables and HER2-positive status in the training cohort, including Ki-67 expression (odds ratio [OR], 4.14 [95% CI: 2.16, 7.95]; P < 0.001), HR status (OR, 0.27 [95% CI: 0.15, 0.48]; P < 0.001), and histology type (OR, 0.22 [95% CI: 0.05, 0.96]; P = 0.04). Following multivariable analysis, Ki-67 expression (odds ratio [OR], 2.93 [95% CI: 1.48, 5.84]; P = 0.002) and HR status (OR, 0.40 [95% CI: 0.21, 0.74]; P = 0.004) remained as independent predictors for differentiating HER2-postive and -negative status (Table 3), which were used to develop the clinical model.

Table 3 Univariable and multivariable analyses of the training cohort to assess variables associated with odds of HER2-positive status
Full size table

For task 2, univariable analysis of variables suggested that Ki-67 expression (OR, 0.74 [95% CI: 0.61, 0.91]; P = 0.01) and HR status (OR, 1.36 [95% CI: 1.10, 1.67]; P = 0.02) were significantly associated with odds of HER2-low status in the training cohort and were thus included in the multivariable analysis (Table 4). However, after multivariable analysis, there were no clinicopathologic variables that emerged as independent predictors associated with odds of HER2-low status. (Table 4).

Table 4 Univariable and multivariable analyses of the training cohort to assess variables associated with odds of HER2-low status
Full size table

Model construction and performance

Models for differentiating HER2-positive versus -negative breast cancers

The Extratrees classifier performed better than others in constructing the radiomics signature and ITH signature. The AUCs of radiomics signature were 0.79 (95% CI: 0.73, 0.85), 0.71 (95% CI: 0.60, 0.81), and 0.70 (95% CI: 0.59, 0.82) in the training, validation and test cohorts, respectively. The ITH signature achieved AUCs of 0.81 (95% CI: 0.75, 0.86), 0.81 (95% CI: 0.74, 0.88), and 0.81 (95% CI: 0.74, 0.89) in the training, validation and test cohorts, respectively (Fig. 4a). The description of the performance of each classifier and other diagnostic metrics are shown in Table S4 and S6 (Additional File 1). Additionally, by integrating clinicopathologic variables with the radiomics signature and ITH signature, the combined model achieved improved performance, with AUCs of 0.83 (95% CI: 0.77, 0.88), 0.84 (95% CI: 0.76, 0.91) and 0.83 (95% CI: 0.75, 0.90) in the training, validation, and test cohorts, respectively. Notably, the ITH signature exhibited higher values in the HER2-positive group compared to the HER2-negative group across all cohorts (P < 0.001) (Fig. 4c).

Fig. 4
figure 4

The performance of models for each task. (a) represent AUCs of each model in the training, validation and test cohorts for task (1) (b) represent AUCs of the ITH signature in the training, validation and test cohorts for task (2) AUCs were reported with 95% CI in parentheses. (c) represent the mean ITH signature of HER2-positive tumors was higher than that of HER2-negative (all P < 0.001). (d) illustrate the mean ITH signature of HER2-low tumors was higher than that of HER2-zero (all P < 0.001). P values were calculated using the Mann-Whitney test. The horizontal lines in the boxes represent the medians, and the whiskers represent the minimum and maximum values. Circles represent outliers. ★★★ represents P<0.001. AUC, area under the receiver operating characteristic curve, CI, confidence interval, ITH, intratumoral heterogeneity, HER2, human epidermal growth factor receptor 2

Full size image

Models for differentiating HER2-low versus -zero breast cancers

The development of the clinical model and the combined model was abandoned due to the absence of clinicopathologic variables that were retained as independent predictors, according to the results of the multivariable analysis. The RF classifier performed better than others in the establishment of ITH signature. The ITH signature achieved outstanding performance, with AUCs of 0.94 (95% CI: 0.91, 0.96), 0.93 (95% CI: 0.88, 0.98), and 0.84 (95% CI: 0.77, 0.92) in the training, validation, and test cohorts, respectively (Fig. 4b). The descriptions of the performance of each classifier and additional diagnostic metrics are detailed in Table S5 and S7 (Additional File 1). Moreover, the ITH signature simultaneously exhibited a significant distribution between HER2-low and -zero status across all cohorts (P < 0.001) (Fig. 4d).

Model evaluation and interpretation

The decision curve analysis was utilized to evaluate the clinical net benefit value. For the distinguishment between HER2-positive and -negative status, when the threshold was set at 0.92, 0.65, and 0.64, respectively, the clinical net benefits were higher than zero in the training, validation, and test cohorts (Fig. 5a). For the distinguishment between HER2-low and -zero status, high net benefits were obtained in both the training and validation cohorts. The clinical net benefit was higher than zero when the threshold was set at 0.83 in the test cohort (Fig. 5b). For the ITH signatures, calibration curve analysis demonstrated that the predicted probabilities and the actual probabilities of both tasks had favorable consistency in the training, validation and test cohorts, respectively (Fig. 5c and d). The SHAP analyses exhibited the top ten features of the ITH signatures for each task (Fig. 5e and f). The SHAP bar chart provided global explanations of the ITH signatures by computing the average SHAP absolute values of the habitat radiomics features. The SHAP beeswarm plots indicated each feature’s positive or negative effects on the prediction probability in red and blue for both tasks. The heatmap plots provided the SHAP values of the habitat radiomics features for individuals, displaying the local explanations and each feature’s direction and intensity of impact. For the distinguishment between HER2-positive and -negative status, the hemodynamic heterogeneity of T1C had a more significant impacts compared to the diffuse heterogeneity of DWI (Fig. 5e). For the differentiation between HER2-low and -zero status, the heterogeneous characteristics of each habitat area in both sequences had significant impacts on the predicted probabilities (Fig. 5f).

Fig. 5
figure 5

The evaluation and interpretation of the ITH signature for each task. The decision curves of the ITH signature for task 1 across three cohorts (a); The decision curves of the ITH signature for task 2 across three cohorts (b); The calibration curves of the ITH signature for task 1 across three cohorts (c); The calibration curves of the ITH signature for task 1 across three cohorts (d); The SHAP analyses of the ITH signatures for task 1 (e) and task 2 (f); the bar charts illustrated the weight of the top ten important features of the ITH signatures. The horizontal axis represents the average shapley values, while the vertical axis represents the habitat radiomics features; the bees-warm plots revealed the relative importance of habitat radiomics features and illustrate their actual relationships with the prediction outcomes. Each individual is represented by a single dot on each feature flow. The horizontal position of the dot is determined by the SHAP value of that feature, and dots accumulated along each feature row to show density; the heatmap plots displayed the local explanations of features. With individuals delineated along the x-axis and habitats radiomics features along the y-axis, these plots encode SHAP values using a color scale. ITH, intratumoral heterogeneity, SHAP, shapley additive explanations analysis, DCA, decision curve analysis, C, contrast-enhanced T1-weighted sequence, DWI, diffusion-weighted imaging, h, habitat

Full size image

Discussion

Our study constructed the ITH signatures based on habitat radiomics features for differentiating HER2-positive and -negative tumors, and further for differentiating HER2-low and -zero tumors. For both tasks, the ITH signatures showed remarkable performance in the training, validation, and test cohorts, respectively (AUCs, 0.81–0.94).

To our knowledge, this is the first time that quantitative analysis of ITH has been employed to differentiate between various HER2 expressions levels via breast MRI. Several recent studies have established multiple models for identifying HER2 subtypes using MRI findings or classic radiomics, achieving relatively robust performance [14,15,16, 23]. However, features extracted from whole-tumor and peritumoral areas fail to capture the nuances of ITH. In fact, predictive models based on habitat or subregion radiomics have demonstrated outstanding performance across various cancers, including esophageal cancer, colorectal cancer, and glioblastoma [24,25,26]. Additionally, Shi et al. [12] utilized quantitative evaluation of ITH to predict treatment response to neoadjuvant chemotherapy in breast cancer, developing a model that incorporated intratumoral ecological diversity features, clinicopathologic variables and C-radiomics, which outperformed models using C-radiomics alone. In our study, ITH signatures were captured from MRI, which is crucial for diagnosing and evaluating breast cancer. The performance of the ITH signature was comparable to that of the combined model integrating clinicopathologic variables, radiomics signatures, and ITH signatures in task 1, while it continued to demonstrate superior performance in task 2. This suggests our model can effectively leverage intratumoral habitats to extract heterogeneous radiomics features from MR images, accurately differentiating HER2 expression levels. This approach could offer valuable insights for clinical decision-making on tailored treatment strategies and therapeutic management.

This study adopted two tasks to initially differentiate between HER2-positive and -negative breast cancers, followed by further distinguishment between HER2-low and -zero cases. Indeed, discrimination between HER2-low and zero tumors is crucial, as lower agreement and accuracy among pathologists have been note when interpreting scanned slides of HER2 1 + and HER2 0 scoring in breast cancer biopsies [27]. Various classification methods have been utilized in recent studies to address this ternary challenge. Zheng et al. [16] developed three different prediction models for distinguishing HER2-positive, -low and -zero from others, respectively. Although each model demonstrated robust performance, this classification process required multiple iterations before a particular subtype was finally defined. Ramtohul et al. [14] and Peng et al. [23] employed a relatively appropriate process to first differentiate HER2-low/-positive versus -negative, and then further differentiating between HER2-low and -positive tumors. However, they still did not directly investigate the differences in imaging phenotypes and clinicopathologic characteristics between HER2-low and -zero tumors. Bian et al. [15] conduced two tasks similar to ours; however, their models relied solely on classic radiomics features for construction. Notably, our signatures, derived from a larger sample size and encompassing more diverse radiomics features from intratumoral habitats, achieved improved performance relative to their findings (their AUCs: 0.71–0.82, ours: 0.81–0.94).

In our univariable logistic regression analysis, HR status, Ki-67 expression, and histological grade were significantly associated with HER2-low tumors. Higher HR status was observed in HER2-low compared to HER2-zero breast cancer, along with lower Ki-67 levels and lower histological grades, consistent with previous reports [28, 29]. However, these clinicopathologic variables were not retained as independent predictors in our subsequent multivariate analysis. This may be attributed to differences in HR status and Ki-67 expression prevalence between two centers. Another possible explanation might be that the characteristic of ITH is more pronounced in HER2-low tumors. HER2 heterogeneity is a significant issue that warrants attention, as it becomes evident when a single tumor contains multiple HER2 amplification patterns, highlighting the heterogeneous patterns of genetic and HER2 protein expression. It has been proven that patients with HER2-low expression may experience different treatment resistances and survival outcomes, despite having consistent molecular subtypes and other clinical features [30, 31]. Furthermore, in HER2-positive breast cancer, ITH is associated with lower rates of pathological complete response (pCR) in patients receiving T-DM1 and pertuzumab, ultimately leading to the development of resistance and preoperative local progression [32]. Therefore, accurately identifying HER2 expression based on intratumoral heterogeneity prior to treatment is critical. In our study, MRI voxel information within the tumor was further classified into more heterogeneous habitats. The ITH signature quantitatively characterized the heterogeneity of hemodynamics and tissue cell density structure, allowing for more precise identification of HER2-postive and -low breast cancers.

This study had certain limitations. First, the retrospective nature of this study, combined with a small-size external test cohort, which necessitates further validation of model’s generalizability in larger multi-center datasets and prospective studies using reliable and reproducible validation methods, such as k-fold cross-validation and bootstrapping method to account for datasets diversity during the validation process. Second, while these tasks have already a relatively convenient way to identify different HER2 statuses, the development of a ternary classification model could contribute to direct identification of HER2 expressions through more advanced algorithms. Lastly, ITH was determined based on imaging information from multiparametric MRI in this study. The correlation between habitat radiomics features and the underlying biological mechanisms, as well as histopathological features of HER2 expression remains unclear. Comparison between MRI and pathological specimens and integration of multi-omics studies to further understand HER2 heterogeneity would be of interest.

Conclusions

In conclusion, our study developed ITH signatures to quantitatively measure ITH utilizing habitat-based MRI radiomics, achieving outstanding performance among current methods for differentiating HER2-positive and -negative tumors, as well as further identifying HER2-low breast cancers. This may contribute to more precise clinical selection of appropriate populations that benefit from anti-HER2 targeted therapies.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Change history

Abbreviations

HER2:

Human epidermal growth factor receptor 2

ITH:

Intratumoral heterogeneity

AUC:

Area under the receiver operating characteristic curve

IHC:

Immunohistochemistry

FISH:

Fluorescence in situ hybridization

HR:

Hormone receptor

DCE-MRI:

Dynamic contrast-enhanced MRI

DWI:

Diffusion-weighted imaging

PACS:

Picture archiving and communication systems

VOI:

Volume of interest

T1C:

The first contrast-enhanced T1-weighted sequence

SLIC:

Simple linear iterative clustering

CH:

Calinski-Harabasz

LASSO:

Least absolute shrinkage and selection operator

LR:

Logistic Regression

RF:

Random Forest

SHAP:

Shapley additive explanations analysis

CI:

Confidence interval

DCA:

Decision curves analysis

OR:

Odds ratio

pCR:

Pathological complete response

References

  1. Loibl S, Gianni L. HER2-positive breast cancer. Lancet. 2017;389(10087):2415–29.

    Article  CAS  PubMed  Google Scholar 

  2. Fehrenbacher L, Cecchini RS, Geyer CE, Rastogi P, Costantino JP, Atkins JN, et al. NSABP B-47/NRG oncology phase III Randomized Trial comparing adjuvant chemotherapy with or without Trastuzumab in high-risk invasive breast Cancer negative for HER2 by FISH and with IHC 1 + or 2. J Clin Oncol. 2020;38(5):444–53.

    Article  CAS  PubMed  Google Scholar 

  3. de Nonneville A, Houvenaeghel G, Cohen M, Sabiani L, Bannier M, Viret F, et al. Pathological complete response rate and disease-free survival after neoadjuvant chemotherapy in patients with HER2-low and HER2-0 breast cancers. Eur J Cancer. 2022;176:181–8.

    Article  PubMed  Google Scholar 

  4. Hu XE, Yang P, Chen S, Wei G, Yuan L, Yang Z, et al. Clinical and biological heterogeneities in triple-negative breast cancer reveals a non-negligible role of HER2-low. Breast Cancer Res. 2023;25(1):34.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  5. Banerji U, van Herpen CML, Saura C, Thistlethwaite F, Lord S, Moreno V, et al. Trastuzumab duocarmazine in locally advanced and metastatic solid tumours and HER2-expressing breast cancer: a phase 1 dose-escalation and dose-expansion study. Lancet Oncol. 2019;20(8):1124–35.

    Article  CAS  PubMed  Google Scholar 

  6. Modi S, Park H, Murthy RK, Iwata H, Tamura K, Tsurutani J, et al. Antitumor Activity and Safety of Trastuzumab Deruxtecan in patients with HER2-Low-expressing advanced breast Cancer: results from a phase ib study. J Clin Oncol. 2020;38(17):1887–96.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  7. Modi S, Jacot W, Yamashita T, Sohn J, Vidal M, Tokunaga E, et al. Trastuzumab Deruxtecan in previously treated HER2-Low advanced breast Cancer. N Engl J Med. 2022;387(1):9–20.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  8. Wolff AC, Hammond MEH, Allison KH, Harvey BE, Mangu PB, Bartlett JMS, et al. Human epidermal growth factor receptor 2 testing in breast Cancer: American Society of Clinical Oncology/College of American Pathologists Clinical Practice Guideline Focused Update. J Clin Oncol. 2018;36(20):2105–22.

    Article  CAS  PubMed  Google Scholar 

  9. Chen R, Qi Y, Huang Y, Liu W, Yang R, Zhao X, et al. Diagnostic value of core needle biopsy for determining HER2 status in breast cancer, especially in the HER2-low population. Breast Cancer Res Treat. 2023;197(1):189–200.

    Article  CAS  PubMed  Google Scholar 

  10. Zardavas D, Irrthum A, Swanton C, Piccart M. Clinical management of breast cancer heterogeneity. Nat Rev Clin Oncol. 2015;12(7):381–94.

    Article  CAS  PubMed  Google Scholar 

  11. Miglietta F, Griguolo G, Bottosso M, Giarratano T, Lo Mele M, Fassan M, et al. Evolution of HER2-low expression from primary to recurrent breast cancer. NPJ Breast Cancer. 2021;7(1):137.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  12. Shi Z, Huang X, Cheng Z, Xu Z, Lin H, Liu C, et al. MRI-based quantification of Intratumoral Heterogeneity for Predicting Treatment response to neoadjuvant chemotherapy in breast Cancer. Radiology. 2023;308(1):e222830.

    Article  PubMed  Google Scholar 

  13. Cho HH, Kim H, Nam SY, Lee JE, Han BK, Ko EY, et al. Measurement of Perfusion heterogeneity within Tumor habitats on Magnetic Resonance Imaging and its association with prognosis in breast Cancer patients. Cancers (Basel). 2022;14(8):1858.

    Article  PubMed  Google Scholar 

  14. Ramtohul T, Djerroudi L, Lissavalid E, Nhy C, Redon L, Ikni L, et al. Multiparametric MRI and Radiomics for the prediction of HER2-Zero, -Low, and -positive breast cancers. Radiology. 2023;308(2):e222646.

    Article  PubMed  Google Scholar 

  15. Bian X, Du S, Yue Z, Gao S, Zhao R, Huang G, et al. Potential antihuman epidermal growth factor receptor 2 target therapy beneficiaries: the role of MRI-Based Radiomics in distinguishing human epidermal growth factor receptor 2-Low status of breast Cancer. J Magn Reson Imaging. 2023;58(5):1603–14.

    Article  PubMed  Google Scholar 

  16. Zheng S, Yang Z, Du G, Zhang Y, Jiang C, Xu T, et al. Discrimination between HER2-overexpressing, -low-expressing, and -zero-expressing statuses in breast cancer using multiparametric MRI-based radiomics. Eur Radiol. 2024;34(9):6132–44.

    Article  CAS  PubMed  Google Scholar 

  17. Gillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures, they are data. Radiology. 2016;278(2):563–77.

    Article  PubMed  Google Scholar 

  18. Achanta R, Shaji A, Smith K, Lucchi A, Fua P, Süsstrunk S. SLIC superpixels compared to state-of-the-art superpixel methods. IEEE Trans Pattern Anal Mach Intell. 2012;34(11):2274–82.

    Article  PubMed  Google Scholar 

  19. Liu Y, Li Z, Xiong H, Gao X, Wu J, Wu S. Understanding and enhancement of internal clustering validation measures. IEEE Trans Cybern. 2013;43(3):982–94.

    Article  PubMed  Google Scholar 

  20. van Griethuysen JJM, Fedorov A, Parmar C, Hosny A, Aucoin N, Narayan V, et al. Computational Radiomics System to Decode the Radiographic phenotype. Cancer Res. 2017;77(21):e104–7.

    Article  PubMed  PubMed Central  Google Scholar 

  21. Zwanenburg A, Vallières M, Abdalah MA, Aerts HJWL, Andrearczyk V, Apte A, et al. The image Biomarker Standardization Initiative: standardized quantitative Radiomics for High-Throughput Image-based phenotyping. Radiology. 2020;295(2):328–38.

    Article  PubMed  Google Scholar 

  22. Nohara Y, Matsumoto K, Soejima H, Nakashima N. Explanation of machine learning models using shapley additive explanation and application for real data in hospital. Comput Methods Programs Biomed. 2022;214:106584.

    Article  PubMed  Google Scholar 

  23. Peng Y, Zhang X, Qiu Y, Li B, Yang Z, Huang J, et al. Development and Validation of MRI Radiomics models to differentiate HER2-Zero, -Low, and -positive breast Cancer. AJR Am J Roentgenol. 2024;222(4):e2330603.

    Article  PubMed  Google Scholar 

  24. Verma R, Correa R, Hill VB, Statsevych V, Bera K, Beig N, et al. Tumor Habitat-derived Radiomic features at pretreatment MRI that are prognostic for progression-free survival in Glioblastoma Are Associated with Key Morphologic attributes at histopathologic examination: a feasibility study. Radiol Artif Intell. 2020;2(6):e190168.

    Article  PubMed  PubMed Central  Google Scholar 

  25. Xie C, Yang P, Zhang X, Xu L, Wang X, Li X, et al. Sub-region based radiomics analysis for survival prediction in oesophageal tumours treated by definitive concurrent chemoradiotherapy. EBioMedicine. 2019;44:289–97.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  26. Cai Z, Xu Z, Chen Y, Zhang R, Guo B, Chen H, et al. Multiparametric MRI subregion radiomics for preoperative assessment of high-risk subregions in microsatellite instability of rectal cancer patients: a multicenter study. Int J Surg. 2024;110(7):4310–9.

    PubMed  PubMed Central  Google Scholar 

  27. Fernandez AI, Liu M, Bellizzi A, Brock J, Fadare O, Hanley K, et al. Examination of low ERBB2 protein expression in breast Cancer tissue. JAMA Oncol. 2022;8(4):1–4.

    Article  PubMed  PubMed Central  Google Scholar 

  28. Denkert C, Seither F, Schneeweiss A, Link T, Blohmer JU, Just M, et al. Clinical and molecular characteristics of HER2-low-positive breast cancer: pooled analysis of individual patient data from four prospective, neoadjuvant clinical trials. Lancet Oncol. 2021;22(8):1151–61.

    Article  CAS  PubMed  Google Scholar 

  29. Won HS, Ahn J, Kim Y, Kim JS, Song JY, Kim HK, et al. Clinical significance of HER2-low expression in early breast cancer: a nationwide study from the Korean breast Cancer Society. Breast Cancer Res. 2022;24(1):22.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  30. Tarantino P, Hamilton E, Tolaney SM, Cortes J, Morganti S, Ferraro E, et al. HER2-Low breast Cancer: pathological and clinical Landscape. J Clin Oncol. 2020;38(17):1951–62.

    Article  CAS  PubMed  Google Scholar 

  31. Li Y, Tsang JY, Tam F, Loong T, Tse GM. Comprehensive characterization of HER2-low breast cancers: implications in prognosis and treatment. EBioMedicine. 2023;91:104571.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  32. Filho OM, Viale G, Stein S, Trippa L, Yardley DA, Mayer IA, et al. Impact of HER2 heterogeneity on treatment response of early-stage HER2-Positive breast Cancer: phase II neoadjuvant clinical trial of T-DM1 combined with Pertuzumab. Cancer Discov. 2021;11(10):2474–87.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

Download references

Acknowledgements

None.

Funding

This work was supported by Peking University People’s Hospital Scientific Research Development Funds [grant numbers RDGS2022-10].

Author information

Author notes

  1. Haoquan Chen and Yulu Liu are co-first-authors of this manuscript.

  2. Shu Wang and Yi Wang are co-corresponding-authors of this manuscript.

Authors and Affiliations

  1. Department of Radiology, Peking University People’s Hospital, No. 11, Xizhimen South St, Beijing, 100044, China

    Haoquan Chen, Yulu Liu, Xiaoxuan Jia, Fan Chai, Nan Hong & Yi Wang

  2. Department of Radiology, The Affiliated Hospital of Southwest Jiaotong University/ The Third People’s Hospital of Chengdu, Chengdu, 610031, China

    Yulu Liu

  3. Department of Radiology, Jiangmen Central Hospital, Jiangmen, 529030, China

    Jiaqi Zhao

  4. Department of Breast Surgery, Peking University People’s Hospital, No. 11, Xizhimen South St, Beijing, 100044, China

    Yuan Peng & Shu Wang

Authors
  1. Haoquan Chen
    View author publications

    You can also search for this author inPubMed Google Scholar

  2. Yulu Liu
    View author publications

    You can also search for this author inPubMed Google Scholar

  3. Jiaqi Zhao
    View author publications

    You can also search for this author inPubMed Google Scholar

  4. Xiaoxuan Jia
    View author publications

    You can also search for this author inPubMed Google Scholar

  5. Fan Chai
    View author publications

    You can also search for this author inPubMed Google Scholar

  6. Yuan Peng
    View author publications

    You can also search for this author inPubMed Google Scholar

  7. Nan Hong
    View author publications

    You can also search for this author inPubMed Google Scholar

  8. Shu Wang
    View author publications

    You can also search for this author inPubMed Google Scholar

  9. Yi Wang
    View author publications

    You can also search for this author inPubMed Google Scholar

Contributions

HQC contributed to the literature search, Figs. 1, 2 and 3, conception design, collection and analysis of data, data interpretation and manuscript writing. YLL contributed to the literature search, Figs. 4 and 5, collection and analysis of data, data interpretation and manuscript writing. JQZ contributed to the provision, collection and analysis of data and data interpretation. XXJ and FC contributed to the data proofreading and analysis. YP contributed to the collection and proofreading of data. NH, SW and YW contributed to the administrative support and manuscript revision. WS and YW contributed to the conception design and funding acquisition. HQC and YLL have directly accessed and verified the underlying data reported in the manuscript. YW was responsible for the decision to submit the manuscript. All authors reviewed the final manuscript, believed it represents valid work and approved the submitted version.

Corresponding authors

Correspondence to Shu Wang or Yi Wang.

Ethics declarations

Ethics approval and consent to participate

This retrospective multicenter study was approved by the institutional review board and ethics committee for review of Peking University People’s Hospital.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

The original version of this article has been revised: the Data Availability section has been updated.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary Material 1

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Chen, H., Liu, Y., Zhao, J. et al. Quantification of intratumoral heterogeneity using habitat-based MRI radiomics to identify HER2-positive, -low and -zero breast cancers: a multicenter study. Breast Cancer Res 26, 160 (2024). https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s13058-024-01921-7

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s13058-024-01921-7

Keywords

Breast Cancer Research

ISSN: 1465-542X

Contact us