MRI-based evaluation for predicting pathological response and monitoring reactive angioma evolution in triple-negative breast cancer undergoing neoadjuvant chemoimmunotherapy Authors: Luyi Lin, Jinlong Zheng, Fan Yang, Ke Zuo, Jiayin Zhou, Guanhua Su, Xiaowen Ma, Chao You, Yajia Gu Published: 2025 ABSTRACT BackgroundNeoadjuvant chemotherapy combined with anti-programmed cell death protein-1 (PD-1) immunomodulators (NACI) has been proven to improve outcomes in triple-negative breast cancer (TNBC). However, the role of MRI in predicting therapeutic response remains underexplored. This study aims to investigate the utility of MRI for early response prediction and longitudinal assessment of reactive angiomatous changes in TNBC patients undergoing NACI therapy.MethodsThis secondary analysis pooled data from the CamRelief, NeoTENNIS, and other randomized controlled trials (RCTs) investigating NACI in our center between January 2020 and December 2023. Data collected included baseline clinical-pathological characteristics, MRI features, and histopathological response outcomes. MRI evaluation incorporated both BI-RADS descriptors and quantitative apparent diffusion coefficient (ADC) histogram metrics derived from whole-tumor segmentation. Univariate and multivariate two-step logistic regression analyses identified independent predictors of pathological complete response (pCR). A prognostic model integrating significant MRI and histopathological variables was developed. Longitudinal reactive angioma evolution during NACI were analyzed by non-parametric test and chi-square test.ResultsA total of 61 TNBC patients undergoing NACI were enrolled. Univariate analysis identified four independent predictors of pCR. While elevated Ki-67 index (p < 0.001) and CD8-T cell infiltration (p = 0.004) reflected baseline pathological predictors, washout kinetic curve (p = 0.004) and reduced median_ADC value (p = 0.021) of MRI provided unique insights with pCR. Moreover, the combined multi-dimensional model achieved a high prediction accuracy, got AUC of 0.89. Serial MRI monitoring identified transient reactive changes characterized by multiple small masses emerging (mean 8.82 ± 0.67 mm) during NACI unrelated to response, peaking at post-cycle-2 MRI, followed by progressive regression. One was histopathologically confirmed as benign reactive angiomatous remodeling.ConclusionMRI demonstrated dual clinical utility in NACI management of TNBC. The integrative model combining MRI biomarkers with clinicopathological features demonstrated superior predictive accuracy for pCR, while reactive angiomatous remodeling exhibited a transient change.Supplementary InformationThe online version contains supplementary material available at 10.1186/s40644-025-00980-z. Background TNBC is a type of breast cancer in which ER, PR, and human epidermal growth factor receptor2 (Her2) are all negative. The disease progresses rapidly and has a poor prognosis. Neoadjuvant therapy is an important treatment for TNBC, while immunotherapy, particularly anti-PD-1/PD-L1 pathway immune modulators, has gained increasing attention. In recent years, several clinical trials of NACI using anti PD-1/PD-L1 immune modulators pathway have achieved good therapeutic effects [1]. However, some TNBC patients still fail to achieve pathological complete response (pCR) after NACI treatment. In addition to systemic adverse events (like pneumonitis) associated with anti-PD-1/PD-L1 immunotherapy in breast cancer, localized adverse events within the breast tissue itself may also occur [2]. Therefore, it is necessary to investigate more accurate and comprehensive biomarkers for therapeutic efficacy prediction and treatment responses assessment. MRI is an important tool for the diagnosis and response evaluation of breast cancer, and its value in pCR predicting in neoadjuvant chemotherapy for TNBC has been recognized [3]. The DCE-MRI based model shows some predictive capability for NACI response in TNBC patients [4]. However, this study lacks exploration of other MRI sequences features as well as microscopic pathological characteristics. At the microscopic level, TNBC could be further classified into four transcriptome-based subtypes: luminal androgen receptor (LAR), immunomodulatory (IM), basal-like immune-suppressed (BLIS), and mesenchymal-like (MES) [5]. Precision therapies tailored to these subtypes, particularly immunotherapy for IM-TNBC, have improved advanced TNBC survival [6–8]. Previous studies have refined the transcriptome-based classification of TNBC into immunohistochemical (IHC)- based subtypes that are more clinically actionable, revealing that IHC markers such as LAR, CD8, and FOXC1 exhibit strong correlations with specific TNBC subtypes [9]. We hypothesize that integrating MRI features with clinicopathological and IHC subtype data could better predict pCR in TNBC patients undergoing NACI. During immunotherapy, the breast may exhibit unique adverse events such as pseudoprogression, hyperprogression, and mixed response [10]. The emergence of new enhancement lesions in the breast during immunotherapy creates challenges for accurate response evaluation. The CamRelief (NCT04613674) and NeoTENNIS (NCT04418154) RCTs, published in JAMA (December 2024) and eClinicalMedicine (August 2024), respectively, demonstrated that combined locally developed PD-1 immune checkpoint inhibitors developed in China (camrelizumab or toripalimab) with neoadjuvant chemotherapy significantly improved pCR rates in early/locally advanced TNBC [11, 12]. Our hospital was a key participating site and the corresponding affiliation in these studies. Based on these well-designed RCT cohorts and matched longitudinal MRI data, comprehensive efficacy evaluation and adverse event monitoring can be anticipated. Therefore, based on the secondary analysis of CamRelief and NeoTENNIS and other RCTs, this article aims at constructing a multi-dimensional pCR prediction model including MRI and clinicopathological features, and longitudinally monitoring new enhancement lesions changes in breast during NACT treatment. Methods Subjects The patient recruitment of this study was presented in Fig. 1. This research is a secondary analysis of CamRelief, NeoTENNIS and other TNBC-NACI RCTs. The study was approved by the Ethics Committee of the Institutional Review Board (IRB) of our institution. All patients provided written informed consent before enrollment. In detail, we collected patients diagnosed as TNBC in our hospital from January 2020 to December 2023. The inclusion criteria included: (I) no treatment was conducted in other hospitals previously; (II) MRI scans performed before core-needle biopsy; (III) participated in the relevant RCTs of our hospital and received NACI treatment. The exclusion criteria were: (I) poor quality of MRI scans; (II) NACI treatment was not completed. All patients underwent biopsy and IHC test (traditional and TNBC-subtype relevant) at the time of baseline, and some patients underwent transcriptome-based subtype test. During the NACI treatment, MRI scans were performed every two cycles. The clinical and pathological data at baseline of these patients (n = 66), all MRI data at baseline and during NACI treatment, and the pathological data of surgical specimens were collected. Fig. 1Patient recruitment of this secondary analysis research of CamRelief, NeoTENNIS and other TNBC-NACI RCTs Patient recruitment of this secondary analysis research of CamRelief, NeoTENNIS and other TNBC-NACI RCTs Clinical and pathological data The clinical information included age, menopausal status, pre-NACI T stage, lymph node metastases status and NACI regimen. The traditional pathological information included Ki-67 index before NACI and the pCR status evaluation after surgery. The TNBC-subtype relevant IHC data at baseline were collected from pathology reports: percentage of positive cells for CD8, AR, FOXC1. According to the requirements of RCTs, patients who underwent transcriptomic or digital pathological subtype test were classified according to their reports. pCR was defined as the absence of invasive cancer burden in either the breast or associated axillary lymph nodes (ypT0/is ypN0). MRI technique and features At our center, scanning was performed on three types of scanners: a Siemens 3.0-T MRI scanner (Siemens Healthineers, Erlangen, Germany), and a GE 1.5-T MRI scanner (GE, Signa HDx) with a 16-channel body coil. Details were shown in the affiliated file. Based on the BI-RADS and results from other literature studies, two radiologists evaluated conventional tumor imaging features in MRI, including background parenchymal enhancement (BPE), delayed phase of kinetic curve assessment, lesion type, length, peritumoral edema, and others. Additionally, apparent diffusion coefficient (ADC) histogram features were analyzed. The tumors were manually delineated on the high-b value DWI by a radiologist, to generate three-dimensional regions of interest (ROIs) using ITK-SNAP software (version 3.8.0). Tumor areas were defined as hyperintensity on DW images by avoiding T2 shine-through regions (e.g., cystic and necrotic components). DCE MRI was used for lesion localization and boundary verification [13]. The ADC value was calculated using a monoexponential model based on the following equation [14]:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:S={S}_{0}\mathrm{*}\mathrm{e}\mathrm{x}\mathrm{p}(-b*ADC)$$\end{document} Where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:S$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{S}_{0}$$\end{document} denote the signal intensities in the presence and absence of diffusion sensitization, respectively, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$b$$\end{document} represents the diffusion sensitization factor, which determines the degree of diffusion weighting applied to the signal intensity. Histogram-based features, including 5th and 95th percentiles of ADC values, mean, median, standard deviation, variance, kurtosis(a measure of the shape of the probability distribution), skewness(a measure of asymmetry of the probability distribution), energy(a measure of histogram compactness based on squared normalized frequencies), contrast(a measure of the signal difference) and entropy(a measure of texture irregularity), were extracted on the parametric maps within 3D ROIs using Python (version 3.7). New enhancement lesion during NACI New enhancement lesions were defined as newly appearing, significantly enhancing lesions on during-treatment DCE-MRI that were distinct from primary tumor lesions when compared to baseline MRI. These lesions were not restricted by laterality (ipsilateral or contralateral to the primary tumor) or spatial relationship to the tumor lesion at baseline. For clinical practicality and to ensure measurement reliability, only lesions with max length >5 mm at during treatment MRI were included in the final analysis. Statistical analysis The data analyses and processes were implemented with SPSS (version 26.0) and R software (version 4.2.2). Continuous variables were compared by two-sample t tests or non-parametric test, while qualitative variables were analyzed by the chi-square test or Fisher’s exact test. For all tests, P < 0.05 was considered to indicate statistical significance. Univariate analysis and multivariate logistic regression were used to select key features to predict pCR. Receiver operating characteristic curves (ROC) were used to evaluate the different models. The AUC with 95% confidence interval (CI), decision curve analysis (DCA) and calibration curve were calculated. Nomogram were drawn based on the model with the highest AUC. Results Relations between clinical, pathological features, TNBC-subtype related IHC characteristics and NACI response There are 61 TNBC patients completed NACI treatment collected in this research. Of all the 61 TNBC patients treated with NACI, compare between non-pCR and pCR groups, we found that Ki67 index was significantly correlated with NACI response. Patients with high Ki67 index tumor were more likely to get pCR after NACI (p < 0.001). Age, menopausal status, T stage, lymph node metastases were not found significantly difference. In terms of treatment regimens, there was no significant difference in the distribution of non-anthracycline based and anthracycline based regimen between the non-pCR and pCR groups (Table 1). In this study, 60 patients underwent TNBC-subtype test (1 missed), including 23 cases (37.7%) of TNBC-LAR subtype, 31 cases (50.82%) of TNBC-IM subtype, and 6 cases (9.84%) of TNBC-BLIS subtype. There was no significant difference in the distribution of TNBC-subtype between the non-pCR and pCR groups after NACI. Neither TNBC-IM or TNBC-none IM subtype tumors were found to be significantly associated with NACI response. However, by analyzing the characteristics of TNBC-subtype, we found that high CD8 positive cell percentage patients were more likely to get pCR after NACI treatment (p = 0.004) in univariate analysis. The clinical, pathological and TNBC-subtype features with p < 0.1 were included in the multivariate logistic analysis, and Ki67 index and CD8 positive cells proportion had independent predictive value (AUC = 0.84). The higher the Ki67 index(OR = 1.07, p =<0.001) and CD8 positive cells proportion (OR = 1.13, p = 0.011) tended to achieve pCR, as shown in Table 3. Table 1Differences of clinical and pathological features between non-pCR and pCR groups of NACI treatmentnon-pCR(N = 33)pCR(N = 28) P Age (year, mean ± SE)0.58845.36 ± 1.8543.75 ± 2.37Menopausal status (N, %)0.58 Pre-20, 60.6115, 53.57 Post-13, 39.3913, 46.43T stage (N, %)0.686 T2 20, 60.6118, 64.29 T33, 9.091, 3.57 T410, 30.309, 32.14Lymph node metastases status (N, %)0.195 Negative10, 30.3013, 46.43 Positive23, 69.7015, 53.57Ki67 Index(mean ± SE)<0.001*53.18 ± 3.4773.18 ± 3.33NAC regimen (N, %)0.171 Non-anthracycline based8, 24.243, 10.71 Anthracycline based25, 75.7625, 89.29CD8 positive cells (%, mean ± SE)0.004*11.78 ± 1.2419.54 ± 2.42FOXC1 positive cells (%, mean ± SE)0.63848.69 ± 7.2453.39 ± 7.07AR positive cells (%, mean ± SE)0.16633.75 ± 6.9916.89 ± 6.13 Table 1Differences of clinical and pathological features between non-pCR and pCR groups of NACI treatmentnon-pCR(N = 33)pCR(N = 28) P Age (year, mean ± SE)0.58845.36 ± 1.8543.75 ± 2.37Menopausal status (N, %)0.58 Pre-20, 60.6115, 53.57 Post-13, 39.3913, 46.43T stage (N, %)0.686 T2 20, 60.6118, 64.29 T33, 9.091, 3.57 T410, 30.309, 32.14Lymph node metastases status (N, %)0.195 Negative10, 30.3013, 46.43 Positive23, 69.7015, 53.57Ki67 Index(mean ± SE)<0.001*53.18 ± 3.4773.18 ± 3.33NAC regimen (N, %)0.171 Non-anthracycline based8, 24.243, 10.71 Anthracycline based25, 75.7625, 89.29CD8 positive cells (%, mean ± SE)0.004*11.78 ± 1.2419.54 ± 2.42FOXC1 positive cells (%, mean ± SE)0.63848.69 ± 7.2453.39 ± 7.07AR positive cells (%, mean ± SE)0.16633.75 ± 6.9916.89 ± 6.13 Differences of clinical and pathological features between non-pCR and pCR groups of NACI treatment Relations between MRI features and NACI response By analyzing the MRI features of the baseline breast MRI scans, we found that the lesions with high AR positive ratio (> 10%) were more likely to present as T2WI hyperintense within the tumor and washout kinetic curve (p = 0.033). The lesions with high FOXC1 positive ratio (> 10%) had a lower ADC value (1.047 ± 0.124 *10− 3 vs. 1.141 ± 0.179 *10− 3, p = 0.026). Correlation analysis between MRI features and NACI response revealed several significant associations (Table 2). Patients achieving pCR demonstrated significantly lower mean and median ADC values compared to non-pCR patients (p = 0.025 and p = 0.026, respectively). Tumors exhibiting wash out kinetic curve showed reduced pCR likelihood (p = 0.004), while concentric shrinkage (CS) pattern observed at post-C2 MRI was associated with higher pCR rates (p = 0.010). Multivariate logistic regression incorporating imaging features with p < 0.10 demonstrated the MRI model’s predictive capacity for pCR (AUC = 0.76). As presented in Table 3, lower median_ADC (OR = 0.38, p = 0.009) and washout kinetic curve (OR = 5.29, p = 0.007) emerged as independent predictors of NACI response. Table 2Differences of MRI features between non-pCR and pCR groups of NACI treatmentnon-pCR (n = 33)pCR (n = 28) P Conventional MRI feature BPE (N, %)0.202 Minimal + mild28, 84.8520, 71.43 Moderate + marked5, 15.158, 28.57Length (mean ± SE)0.05341.91 ± 3.0234.50 ± 2.22Lesion type (N, %)0.141 Mass25, 75.6726, 92.86 Non-mass enhancement (NME)5, 15.152, 7.14 Mass + NME3, 9.090, 0.00Peritumoral edema (N, %)0.432 Without10, 30.306, 21.43 With23, 69.7022, 78.57kinetic curve type (N, %)0.004* Washout7, 21.2016, 57.1 Not-washout26, 78.812, 42.9Shrikage pattern (N, %) Concentric shrinkage (CS)14, 42.4221, 75.000.010* Non-CS19, 57.887, 25.00 Quantitative ADC Histogram features (*10^-3) Mean (mean ± SE)1.13 ± 0.031.03 ± 0.020.025*Median (mean ± SE)1.10 ± 0.030.99 ± 0.020.021*Standard deviationn (mean ± SE)0.27 ± 0.010.25 ± 0.070.241 Table 2Differences of MRI features between non-pCR and pCR groups of NACI treatmentnon-pCR (n = 33)pCR (n = 28) P Conventional MRI feature BPE (N, %)0.202 Minimal + mild28, 84.8520, 71.43 Moderate + marked5, 15.158, 28.57Length (mean ± SE)0.05341.91 ± 3.0234.50 ± 2.22Lesion type (N, %)0.141 Mass25, 75.6726, 92.86 Non-mass enhancement (NME)5, 15.152, 7.14 Mass + NME3, 9.090, 0.00Peritumoral edema (N, %)0.432 Without10, 30.306, 21.43 With23, 69.7022, 78.57kinetic curve type (N, %)0.004* Washout7, 21.2016, 57.1 Not-washout26, 78.812, 42.9Shrikage pattern (N, %) Concentric shrinkage (CS)14, 42.4221, 75.000.010* Non-CS19, 57.887, 25.00 Quantitative ADC Histogram features (*10^-3) Mean (mean ± SE)1.13 ± 0.031.03 ± 0.020.025*Median (mean ± SE)1.10 ± 0.030.99 ± 0.020.021*Standard deviationn (mean ± SE)0.27 ± 0.010.25 ± 0.070.241 Differences of MRI features between non-pCR and pCR groups of NACI treatment Value of multi-dimensional feature fusion in predicting efficacy of NACI response As shown in Table 3, we attempted to fuse the features of the above several parts. The multivariate logistic regression indicated that Ki67 index, CD8 positive cells proportion, kinetic curve, and median_ADC had significantly predictive value for NACI response (AUC = 0.89). The combined model got the highest AUC than clinicopathological model and MRI model, as shown in Fig. 2(a). The accuracy, sensitivity, and specificity for the combined model were 0.80, 0.71, 0.86. Figure 2(b-d) also show the DCA, calibration curve and nomogram of the combined model. However, the Delong test found that the predictive efficiency of the fusion model was not significantly higher than that of another model. Table 3Prediction model for pCR after NACI treatmentOR (95%CI) P AUC Model1: clinicopathological model 0.84 Ki671.07 (1.03–1.11)<0.001 CD8 positive cell (%)1.13 (1.04–1.26)0.011 Model2: MRI model 0.76 Kinetic curve (washout/not-washout)5.29 (1.66–19.05)0.007 Median_ADC0.38 (0.17–0.74)0.009 Model3: combined model 0.89 Ki67// CD8 positive cell (%)// Kinetic curve (washout/not-washout)// Median_ADC//#1 and #2: Statistically significantly different features between pCR/non-pCR groups were included in the multivariate analysis. Predictive model was constructed by selected features#3: The features selected from the above two models were integrated to directly construct a multidimensional feature model Table 3Prediction model for pCR after NACI treatmentOR (95%CI) P AUC Model1: clinicopathological model 0.84 Ki671.07 (1.03–1.11)<0.001 CD8 positive cell (%)1.13 (1.04–1.26)0.011 Model2: MRI model 0.76 Kinetic curve (washout/not-washout)5.29 (1.66–19.05)0.007 Median_ADC0.38 (0.17–0.74)0.009 Model3: combined model 0.89 Ki67// CD8 positive cell (%)// Kinetic curve (washout/not-washout)// Median_ADC//#1 and #2: Statistically significantly different features between pCR/non-pCR groups were included in the multivariate analysis. Predictive model was constructed by selected features#3: The features selected from the above two models were integrated to directly construct a multidimensional feature model Prediction model for pCR after NACI treatment #1 and #2: Statistically significantly different features between pCR/non-pCR groups were included in the multivariate analysis. Predictive model was constructed by selected features #3: The features selected from the above two models were integrated to directly construct a multidimensional feature model Fig. 2Multi-dimensional feature fusion was expected to predict pCR after NACI treatment in TNBC patients. (a) AUC of the clinicopathological model (model1), MRI feature model (model2) and combined model (model3). Combined model had the highest AUC. (b, c, d) DCA curve, calibration curve and nomogram of the combined model Multi-dimensional feature fusion was expected to predict pCR after NACI treatment in TNBC patients. (a) AUC of the clinicopathological model (model1), MRI feature model (model2) and combined model (model3). Combined model had the highest AUC. (b, c, d) DCA curve, calibration curve and nomogram of the combined model New enhancement lesions during NACI treatment The MRI data of the 61 TNBC patients treated with NACI were analyzed at baseline and every two cycles of follow-up. Of these, all patients underwent breast MRI at baseline and post-C2. Fifty-eight patients underwent breast MRI post-C4. Fifty-five patients completed C6 treatment and subsequently underwent breast MRI. Fifty patients completed C8 treatment and subsequently underwent breast MRI. Among the 61 patients in this study, anti-PD-1 /PD-L1 pathway immune modulators were chosen for NACI treatment with camrelizumab in 25 patients, pembrolizumab in 2 patients, and toripalimab in 34 patients. During the NACI follow-up, we found that some patients had new enhancement lesions on DCE-MRI after the application of immunomodulators. This part summarized the phenomenon. For the convenience of clinical application, only enhancement with a max length > 5 mm were statistically analyzed. As shown in Table 4, 36.07% TNBC patients got new enhancement lesion at post-C2 MRI. Mean number of enhancements was 8.36 ± 1.72, and max length was 8.82 ± 0.67 mm. In MRI Characteristics, the findings were shown as diffusely distributed bilateral breast lesions with marked multifocal enhancement. 86.36% of enhancing foci presented as mass-like lesions; 59.09% exhibited irregular shape; 81.82% of enhancements occurred outside the baseline tumor region. Relations between new enhancement lesions and NACI response MRI features at post-C2 MRI of the new enhancement lesions cases during NACI treatment were analyzed for the association with NACI response. We found that there were no significant differences in number, max length, location, shape, margin, with or without NME between pCR group and non-pCR group (p>0.05). Details were shown in Table 4. This finding revealed that the occurrence of new enhancement lesions during NACI treatment was merely a therapeutic response and did not affect whether pCR was achieved. Table 4Differences of reactive Angioma between non-pCR and pCR groups of NACI TNBC patientsMRI featuresAll(n = 22)Non-pCR(n = 11)pCR(n = 11) P #1 Number (mean ± SD)8.36 ± 1.726.09 ± 3.5910.64 ± 10.620.606Max Length (mm, mean ± SD)8.82 ± 0.677.91 ± 3.089.73 ± 3.100.151Type (N, %)1.000 Mass19, 86.369, 81.8210, 90.91 Mass + NME3, 13.642, 18.181, 9.09Shape#2 (N, %)0.387 Irregular13, 59.095, 45.458, 72.73 No-irregular9, 40.916, 54.553, 27.27Location of tumor range in baseline, (N, %)0.586 Outside18, 81.8210, 90.918, 72.73 Inside and outside4, 18.181, 9.093, 27.27#1: p value means difference between pCR/non-pCR groups#2: whether irregular shape was present within mass lesions Table 4Differences of reactive Angioma between non-pCR and pCR groups of NACI TNBC patientsMRI featuresAll(n = 22)Non-pCR(n = 11)pCR(n = 11) P #1 Number (mean ± SD)8.36 ± 1.726.09 ± 3.5910.64 ± 10.620.606Max Length (mm, mean ± SD)8.82 ± 0.677.91 ± 3.089.73 ± 3.100.151Type (N, %)1.000 Mass19, 86.369, 81.8210, 90.91 Mass + NME3, 13.642, 18.181, 9.09Shape#2 (N, %)0.387 Irregular13, 59.095, 45.458, 72.73 No-irregular9, 40.916, 54.553, 27.27Location of tumor range in baseline, (N, %)0.586 Outside18, 81.8210, 90.918, 72.73 Inside and outside4, 18.181, 9.093, 27.27#1: p value means difference between pCR/non-pCR groups#2: whether irregular shape was present within mass lesions Differences of reactive Angioma between non-pCR and pCR groups of NACI TNBC patients #1: p value means difference between pCR/non-pCR groups #2: whether irregular shape was present within mass lesions Pathologically confirmed reactive angioma Coincidentally, one patient underwent core-needle biopsy of the new enhancement lesion at the original tumor lesion site due to the need of the clinical trial after 2-cycle NACI treatment. The pathological report of this patient showed “angioma”, as shown in Fig. 3. Therefore, we believe that the new enhancement lesion emerged in breast after the application of anti-PD-1/PD-L1 pathway immunomodulators, especially camrelizumab, should be a treatment response rather than disease progression. We call this phenomenon “reactive angioma”. Reactive angioma is not related to NACI response in TNBC. Radiologists need to pay attention to the reactive angioma in clinical work. Fig. 3Representative MRI and H&E staining at baseline and during NACI treatment (camrelizumab). MRI: at baseline (a, T2WI; b, ADC; c, contrast) and during NACI treatment (d-f, post C2, C4, C6). Hematoxylin-eosin (H&E) staining: from post-C2 core needle biopsy specimens (g). The 29-year-old woman with low median_ADC (1.04*10− 3), washout kinetic curve breast cancer lesion got pCR after 6 cycle NACI treatment. Moreover, new enhancement lesion appeared at post-C2, and became smaller at post-C4 and post-C6. Pathological examination of post-C2 core needle biopsy targeting the most prominently enhanced lesion at the primary tumor site revealed a diagnosis of “angioma.” Representative MRI and H&E staining at baseline and during NACI treatment (camrelizumab). MRI: at baseline (a, T2WI; b, ADC; c, contrast) and during NACI treatment (d-f, post C2, C4, C6). Hematoxylin-eosin (H&E) staining: from post-C2 core needle biopsy specimens (g). The 29-year-old woman with low median_ADC (1.04*10− 3), washout kinetic curve breast cancer lesion got pCR after 6 cycle NACI treatment. Moreover, new enhancement lesion appeared at post-C2, and became smaller at post-C4 and post-C6. Pathological examination of post-C2 core needle biopsy targeting the most prominently enhanced lesion at the primary tumor site revealed a diagnosis of “angioma.” Longitudinal assessment of reactive angioma changes during treatment We further analyzed the incidence of reactive angioma across different drug treatment group. For patients treated with camrelizumab, 21 patients (84.00%) showed new enhancement lesions in breast at post-C2. In the patients treated with pembrolizumab or toripalimab, only 1 patient (2.78%) had new enhancement lesions, which was significantly less than that in the camrelizumab group (p < 0.001). Reactive angioma exhibited significantly higher prevalence among camrelizumab-treated TNBC patients, as shown in Fig. 4(a). Further longitudinal analysis of post-C4, post-C6 and post-C8 breast MRI scans of patients treated with camrelizumab was performed. Among them, 1, 3, 7 patients did not undergo post-C4/C6/C8 MRI scan. Patients without reactive angioma at post-C2 consistently excluded in follow-up measurements. Comparing the changes of enhancement lesions in follow-up examinations of camrelizumab treated patients, we found stable lesion counts throughout treatment (post-C2 vs. post-C4/C6/C8: p = 0.420, 0.207, 0.843). However, the max length of enhancement lesions at post-C4, post-C6 and post-C8 scans were significantly smaller than those at post-C2 (p = 0.006, 0.002, 0.004), as shown in Fig. 4(b, c). Fig. 4Reactive angioma incidence and changes during NACI treatment. (a) The incidence of reactive angioma was significantly higher in the camrelizumab-treated group. (b) The number of reactive angioma showed no statistically significant difference across treatment cycles. (c) The maximal length (max length) of reactive angioma peaked at post-C2 and subsequently decreased, with post-C4, C6, and C8 measurements being significantly smaller than those at post-C2 Reactive angioma incidence and changes during NACI treatment. (a) The incidence of reactive angioma was significantly higher in the camrelizumab-treated group. (b) The number of reactive angioma showed no statistically significant difference across treatment cycles. (c) The maximal length (max length) of reactive angioma peaked at post-C2 and subsequently decreased, with post-C4, C6, and C8 measurements being significantly smaller than those at post-C2 Discussion This study developed a multi-dimensional feature prediction model to effectively predict pCR, based on the secondary analysis of the CamRelief, NeoTENNIS and other TNBC-NACI RCTs. The Ki67 index, the percentage of CD8 positive cells, the kinetic curve type, and the median_ADC feature were able to independently predict the response to NACI. In addition, this study longitudinally summarized the phenomenon of new enhancement lesions and found that this phenomenon represented a treatment response (“reactive angioma”) but was unrelated to pCR. It was most obvious on post-C2 MRI and significantly decreased in subsequent treatment cycles. The above findings enhance the understanding of response prediction and adverse event monitoring in NACI treatment and may assist in optimizing of treatment regimens for TNBC patients. Ki67 index is an important indicator of tumor proliferation activity, which indicates poor prognosis, but predicts improved chemotherapy response [15]. This is explained by the fact that highly proliferative tumors respond better to neoadjuvant chemotherapy. Another study based on NACI treatment in TNBC patients reported that proliferation-related gene sets were positively associated with increased pCR, but not with improved prognosis [16]. Moreover, they observed that reduction of proliferation-related gene expression after immunomodulators without combined chemotherapy. It might be caused by immune activation, which might lead to an inhibition of proliferation pathways by the activated immune system [16]. Based on previous studies, we directly demonstrated that high Ki67 index was an independent predictor of pCR in the NACI population. This research discovered that the CD8 positive cell proportion served as an independent predictor of pCR response after NACI treatment in TNBC patients. CD8 + T cells, namely effector T cells, constitute a significant part of the immune system and represent a crucial link in its functioning. Immune inhibitors targeting the PD-1/PD-L1 pathway can activate the patient’s immune system by alleviating the inhibitory effect of this pathway on immune activity, thereby exerting an anti-tumor effect [17, 18]. The quantity of baseline CD8 + cells reflects, to a certain extent, the immune activity of the tumor lesion in the patient. Other researchers also found CD8 + T cells distinctly located at the invasive tumor margin are associated with expression of the PD-1/PD-L1 immune inhibitory axis and may predict response to therapy in metastatic melanoma [19]. Similarly, TNBC patients with a high CD8 + level are more susceptible to immunotherapy and have a greater probability of achieving pCR after NACI treatment. Furthermore, the CD8 positive proportion is closely associated with the TNBC subtype [9]. Patients with a low AR positive proportion and a high CD8 positive proportion are regarded as the TNBC-IM type, and immunotherapy significantly prolongs the survival period of advanced patients of this type [6–8]. Although this study did not directly identify the predictive value of the TNBC subtype for the NACI response, the possible reason might be the insufficiently rich sample size. Nevertheless, the independent predictive value of the CD8 positive proportion also reflects, to a certain extent, the significant role of the TNBC subtype related IHC characteristics in neoadjuvant therapy. There were only a few studies aiming at the correlation of MRI features and pCR response after TNBC NACI treatment. Ramtohul et al. established the response predict model based on DCE-MRI, without indexes of other sequences [4]. Zou et al. explored the relation of pCR and MRI features mainly focusing on tumor shrinkage pattern [20]. This divergence may be attributed to the broader spectrum of MRI features analyzed in our study, as well as the fact that Zou et al. assessed the association between CS and pCR using only the chi-square test. In terms of innovation, our studies included multiple-modal MRI features. Kinetic curve, inflecting the perfusion of tumor, was related with NACI pCR response. Type III enhancement kinetics is recognized as a malignant imaging biomarker. TNBC patients exhibiting this pattern demonstrate higher pCR rates following NACI, potentially attributable to their distinct microcirculation profiles. DWI indexes measuring the mobility of water molecules in tissues and providing information regarding the tissue cellularity are important in breast cancer [21–23]. A few researchers found low pre-treatment ADC was related with pCR in TNBC patients [24]. Moreover, Kennedy et al. demonstrated that low pre-treatment ADC was related with increased tumor-infiltrating lymphocytes (TILs) after trastuzumab treatment in a Her2 positive breast cancer cohort [25]. To our knowledge, there was no study on the relation between ADC and pCR in TNBC NACI treatment patients. ADC correlates with neoadjuvant chemotherapy and immune-activated tumor microenvironments in breast cancer. Median_ADC as a histogram index of ADC, could more accurately reflect the cell density in tumor lesion because of excluding other structures like micro-necrotic areas. Median_ADC could predict the NACI pCR response in TNBC patients, as we demonstrated in this study. We also attempted to integrate multidimensional features, including the Ki67 index, CD8 positive cell proportion, kinetic curve type, and median_ADC, to construct a predictive model for NACI response in TNBC patients. The AUC value of the multi-dimensional feature model was higher than that of the single-dimensional feature model, although the difference was not statistically significant. This finding suggests the potential of integrating clinicopathological, and MRI features to optimize the predictive efficacy of treatment outcomes. We will further in-depth investigate. Reactive cutaneous capillary endothelial proliferation (RCCEP) is common and specific adverse event related to immunotherapy, but lack of comprehensive analyses [26]. In a study of camrelizumab treated advanced hepatocellular carcinoma patients, researchers found RCCEP occurred on the skin surface and was an immune response of skin capillary endothelial cells [27]. In the field of breast cancer, clinical trials have found that the most frequent immune-related adverse event was reactive capillary endothelial proliferation, which reached 87.8% [11]. However, there are few reports about the proliferation of vascular endothelial cells in the breast gland based on DCE-MRI. There is abundant blood flow and dense blood vessels in the mammary gland. Based on core needle biopsy findings from a representative case and above literatures, this study confirmed the presence of a unique reactive angiomatosis phenomenon in the breast following NACI, which was independent of pCR. By characterizing its imaging features and implementing longitudinal monitoring, our findings advance the understanding of this distinct immunotherapy-related adverse event localized to the breast. Therefore, the possibility of reactive angioma should be considered in the response evaluation of NACI treatment of breast cancer when the new enhancement lesions are significantly different from those of the primary tumor. If the max length reduction is observed in the follow-up, the diagnosis of reactive angioma is more supported. The limitations of this study are the small sample size and the preliminary statistical results. We will further investigate the predictive value of these relevant features above in future clinical trials with large samples. Another limitation of our study is the paucity of quantitative features on MRI. After a sufficient sample size is collected, radiomics methods will be used to further expand the variety of MRI multimodal quantitative features. Conclusion In TNBC patients, high Ki67 index, elevated CD8 + T cell ratio, washout kinetic curve, and low median_ADC were independent predictors of pCR. It demonstrated the value of integrating clinicopathological and MRI features in NACI response prediction. Additionally, we observed new enhancement lesions on post-C2 MRI during NACI treatment. These lesions decreased with continued therapy and were unrelated to pCR. Biopsy confirmed these as “angioma,” termed “reactive angioma.” This finding improves accurate response event monitoring during NACI treatment. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1