Tumor flare-like response: A novel imaging phenomenon with predictive significance in triple negative breast cancer patients undergoing neoadjuvant immuno-chemotherapy Authors: Bing Jie Zheng, Cheng Zheng Wang, Ho-Young Song, Wen Juan Xu, Ren Zhi Zhang, Jin Rong Qu, Chun Miao Xu, Xia Yu Yue, Jia Yu Wang, Yu Chen, Ming Ge Liu, Cong Yu Li, Ya Fei Chen, Quan Li Gao, Xue Jun Chen, Xiang Li, Zhen Zhen Liu Published: 2026 ABSTRACT BackgroundPreoperative immuno-chemotherapy improves outcomes in triple-negative breast cancer, but associated imaging response patterns remain poorly characterized. This study aimed to describe a novel MRI phenomenon, tumor flare-like response (TFLR), and evaluate its association with pathological complete response (pCR).MethodsAd hoc imaging analysis of a prospective phase II trial (NCT04213898; n = 39). Breast MRI was performed at baseline and after every two cycles. TFLR was defined a priori as new enhancing nodules (≥5 mm) separate from the index tumor on DCE-MRI, appearing during neoadjuvant therapy with subsequent regression. Inter-reader agreement was assessed with Cohen's kappa and intraclass correlation coefficient. Univariate logistic regression and ROC analysis (Youden index) with 1000-bootstrap resampling identified the optimal largest-nodule size cut-off for pCR prediction. Two multivariable logistic regression models (continuous-size and binary-size) evaluated independence after adjustment for PD-L1 status, FGT category, baseline tumor size, and clinical stage; multicollinearity was assessed by variance inflation factor.ResultsTFLR occurred in 74.4% (29/39) of patients and was significantly associated with higher FGT density (P = .03). Among the 29 patients with TFLR, nodules were predominantly oval (93.1%) with circumscribed margins, homogeneous enhancement. Distribution was bilateral (72.5%), ipsilateral (17.2%), or contralateral (10.3%) relative to the primary tumor, with consistently asymmetric counts between breasts. Kinetic curves were persistent (31.0%), plateau (27.6%), or washout (41.4%). TFLR first appeared after cycle 2 in all cases and completely resolved in 75.9% (22/29) by treatment completion; median persistence was 152 days (95% CI 145–159). Presence of TFLR alone did not predict pCR (P = .72). Patients with largest nodule ≥9 mm had significantly higher pCR rate (77.8% vs. 36.4%, P = .048). ROC analysis identified 9 mm as the optimal cut-off. In multivariable analysis, largest nodule diameter remained independently predictive whether analyzed continuously (OR = 2.965, 95% CI 1.298–6.772, P = .01) or dichotomized at ≥9 mm (OR = 7.833, 95% CI 1.260–48.701, P = .027). Model AUCs were 0.904 (95% CI 0.797-1.00) and 0.727 (95% CI 0.519-0.936), respectively.ConclusionsTFLR is a frequent, reversible MRI finding in TNBC treated with camrelizumab-based neoadjuvant immuno-chemotherapy. Largest nodule diameter ≥9 mm is a strong, independent predictor of pathological complete response and represents a promising early, non-invasive imaging biomarker of immunotherapeutic efficacy. Introduction Triple-negative breast cancer (TNBC), defined by the absence of estrogen receptor, progesterone receptor, and human epidermal growth factor receptor 2 expression, comprises approximately 10-15% of breast cancer cases [1,2]. Its aggressive nature, unfavorable prognosis, and lack of targetable markers pose significant therapeutic challenges [3,4]. Neoadjuvant chemotherapy is the standard treatment for stage II–III TNBC, enabling tumor downstaging and prognostic stratification by pathological complete response (pCR) [[5], [6], [7], [8]]. The advent of immune checkpoint inhibitors has transformed the treatment landscape for early-stage TNBC, demonstrating promising results when combined with chemotherapy [[9], [10], [11], [12]]. However, immune activation frequently induces atypical imaging findings that challenge conventional response criteria. Tumor Flare Reaction (TFR), initially described in other malignancies, manifests as transient tumor enlargement or new lesions reflecting immune-related changes rather than true disease progression [[13], [14], [15]]. Its radiologic correlates remain underexplored in breast cancer patients receiving immune checkpoint inhibitors. Moreover, the understanding of breast MRI imaging in the context of immunotherapy remains incomplete, particularly when considering the alterations within breast tissue induced by immune activation [16]. During neoadjuvant immuno-chemotherapy (NAIC) for TNBC, we observed a distinct breast MRI phenomenon, which we term Tumor Flare-like Response (TFLR). Unlike classic TFR, which mainly reflects primary tumor enlargement, TFLR manifests as new, localized nodules within breast tissue, separate from the primary tumor site. These nodules appear after treatment initiation and regress or resolve on subsequent imaging, which poses challenges for response assessment. This study aimed to characterize TFLR, a previously unreported MRI phenomenon in TNBC, and to and to investigate its association with pCR. Methods Study design and participants This imaging study was an ad hoc analysis of a prospective single-arm, open-label phase II trial (NCT04213898) conducted between April 2020 and May 2024 at a single center. The original trial enrolled 39 women with treatment-naïve, non-metastatic stage II–III TNBC. Key inclusion criteria for the parent trial were age ≥18 years, histologically confirmed TNBC, and clinically significant disease eligible for neoadjuvant therapy (primary-tumor longest diameter ≥2 cm, or 1–2 cm with nodal involvement) per the protocol. Exclusion criteria for the imaging sub-study included MRI contraindications and non-measurable disease on baseline MRI (no lesions >1 cm). The study protocol received institutional review board approval, and written informed consent was obtained from all participants. Primary outcomes including feasibility, efficacy and safety of camrelizumab plus nab-paclitaxel and epirubicin, have been reported elsewhere [17]. The present MRI analysis is reported for the first time. Treatment regimen and response assessment Patients received neoadjuvant treatment consisting of camrelizumab (200 mg, on day 1), nab-paclitaxel (125 mg/m2, on days 1, 8, and 15), and epirubicin (75 mg/m2, on day 1) every three weeks for six cycles. At surgery, pathologic complete response (pCR) was defined as the absence of residual invasive cancer in the breast and sampled axillary nodes, with or without ductal carcinoma in situ. Pathologic response was determined by one breast pathologist (L.M.G., with 20 years of experience). MRI surveillance Participants underwent dynamic contrast-enhanced MRI (DCE-MRI) at baseline and after every two cycles of neoadjuvant therapy on 3-T scanners with dedicated breast coils. Full sequence parameters, including scanner models and DCE-MRI details, are provided in Table S1. Each examination included a high-temporal-resolution dynamic sequence comprising one pre-contrast and 38 post-contrast phases (time frames). Subtraction imaging was applied to all post-contrast phases to evaluate tumor response and detect TFLR. Imaging consistency was maintained through a single-center standardized protocol. Image interpretation Two breast radiologists (X.W.J., 3 years and Z.B.J., 13 years of experience) independently reviewed all MR images using the 2013 Breast Imaging Reporting and Data System (BI-RADS) MR imaging lexicon [18]. Given the novelty of TFLR, consensus criteria were defined prior to formal reading through iterative review and discussion of de-identified study examples (blinded to pCR). The radiologists evaluated a spectrum of characteristics encompassing fibroglandular tissue (FGT), background parenchymal enhancement (BPE), morphologic features of breast tumors, enhancement kinetics, and tumor shrinkage pattern (concentric vs. non-concentric per Fukada et al. [19]). They were blinded to pCR outcomes and treatment response data but had access to basic clinical safety information (e.g., fever, leukocytosis, infection history) to differentiate TFLR from true inflammatory/infectious changes. All timepoints were analyzed sequentially. Discrepancies were resolved by consensus to minimize bias. Documentation of TFLR TFLR was defined a priori at the lesion level as transient enhancing nodules (≥5 mm, per BI-RADS criteria for mass vs. focus on MRI) separate from the index tumor on DCE-MRI, appearing during neoadjuvant therapy and subsequently regressing (Fig. 1). Nodules were distinguished from true tumor progression (persistent growth >20% over consecutive examinations and typically invasive morphology), necrosis (cystic changes with no enhancement), postoperative changes (linear scarring post-biopsy), or infectious/inflammatory mastitis by its characteristic temporal pattern (de novo appearance after cycle 2 followed by regression), absence of corresponding clinical signs of infection (fever, leukocytosis, or mastitis), and review of prior imaging/clinical data to exclude preexisting benign features or concurrent inflammation. For each patient, the number, timing, and duration of TFLR nodules were recorded. Morphologic features and kinetic curves (persistent, plateau, or washout) of TFLR nodules were assessed using the 2013 Breast Imaging Reporting and Data System MR imaging lexicon. Kinetic curves were derived independently by each reader from a small (5–10 mm2) region of interest placed over the most enhancing portion of the largest nodule. For patients with multiple TFLR nodules, only the largest nodule was selected for detailed morphologic and kinetic analysis. Consensus was reached in cases of disagreement between the two radiologists regarding TFLR categorization. TFLR was not described in the initial clinical reports, as this was a retrospective ad hoc analysis.Fig. 1Schematic illustration of tumor flare-like response (TFLR) in TNBC patients during preoperative camrelizumab-based immuno-chemotherapy. TFLR appears as new enhancing nodules separate from the index tumor, emerging during neoadjuvant therapy and regressing or resolving by treatment completion. Schematic illustration of tumor flare-like response (TFLR) in TNBC patients during preoperative camrelizumab-based immuno-chemotherapy. TFLR appears as new enhancing nodules separate from the index tumor, emerging during neoadjuvant therapy and regressing or resolving by treatment completion. Biopsy of TFLR nodules Biopsy of the largest TFLR nodule was performed under targeted ultrasound guidance in four cases where lesions were clearly visible on ultrasound. The target was jointly selected by the breast MRI radiologist who originally identified the TFLR and the interventional radiologist through side-by-side MRI–ultrasound review to ensure imaging–histopathology correlation. The aim was to obtain histologic confirmation of the benign nature of TFLR given the invasive nature of the procedure. Using 16-G core needles (Tru-cut, Bard, Tempe, AZ, USA), biopsies were obtained after cycle 2 (n = 3) or prior to surgery (n = 1). Pathological assessments were conducted by a pathologist with 12 years of experience (C.Y.) and audited by another with 20 years of experience (L.M.G.). After histopathologic confirmation of benign changes, no further biopsies were pursued. Immunohistochemistry was not performed. Statistical analysis Continuous variables were compared using the t-test or Wilcoxon rank-sum test; categorical variables were compared between TFLR and non-TFLR groups and between pCR and non-pCR groups using the χ2 test or Fisher's exact test (cell counts <5). Univariate logistic regression and receiver operating characteristic (ROC) analysis (Youden index) with 1000-bootstrap resampling determined the optimal largest-nodule size cut-off for pCR prediction. Two multivariable logistic regression models were constructed: (1) Continuous-size model, with the largest TFLR nodule diameter entered as a continuous variable; and (2) Binary-size model, using the dichotomized nodule diameter based on the ROC-derived cut-off. Both models were adjusted for initial clinical stage, FGT category, and baseline tumor size. Multicollinearity was assessed by variance inflation factor. Model performance was evaluated by the area under the ROC curve with 95% confidence intervals, sensitivity and specificity were reported at the optimal threshold. The duration of TFLR persistence was estimated by the Kaplan-Meier method. The start point was the date of the first MRI scan demonstrating TFLR, and the end point was the first MRI showing complete disappearance of TFLR. Patients whose TFLR remained visible on the last preoperative MRI were censored at that time point. For patients with multiple nodules, duration was based on the largest nodule. Inter-reader agreement for TFLR presence/absence was assessed using Cohen's kappa. Consistency of maximum diameter measurements of TFLR nodules was evaluated using the intraclass correlation coefficient (ICC, two-way random-effects, absolute agreement) with 95% CI and Bland-Altman analysis. All tests were two-sided, with P < .05 considered statistically significant. Statistical analyses were conducted using IBM SPSS Statistics (version 27.0) and R (version 4.2.1). Results Patients’ characteristics and TFLR incidence The cohort comprised 39 TNBC patients. Median age was 46 years (range 31–59). and median baseline tumor size on MRI was 29.6 mm (range 12.3–56.6). The patients had initial clinical stage II (82.1%; 32 of 39) or III (17.9%; 7 of 39) disease with invasive ductal carcinoma. Twenty-five patients (64.1%) achieved pCR. TFLR occurred in 29 patients (74.4%). Table 1 summarizes baseline characteristics and compares variables between groups. PD-L1 positivity was significantly associated with pCR (P = .01) but not with TFLR occurrence. Patients with TFLR showed significantly higher FGT density (P = .03). No other clinical or imaging parameter differed significantly between groups.Table 1Characteristics of triple-negative breast cancer patients and imaging features associated with tumor flare-like response (TFLR).Table 1CharacteristicAll (n = 39)TFLR StatuspCR StatusYes(n = 29)No(n = 10)P ValuepCR(n = 25)Non-pCR(n = 14)P ValueAge (years)0.110.51 Mean ± SD45 ± 844 ± 849 ± 845 ± 846 ± 9 Median (range)46(31, 59)44(31,58)45(38,59)44 (31, 59)48(33, 59)Longest diameter of breast tumor(mm)0.170.47 Mean ± SD31.3 ± 10.732.7 ± 10.027.2 ± 12.030.3 ± 10.532.9 ± 11.1 Median (range)29.6(12.3,56.6)32.4(12.3,56.6)22.6(16.7,55.3)29.2(12.3,55.3)31.1(19.8,56.6)Menopausal status0.160.65 Premenopausal33(84.6)26(89.7)7(70.0)22(88.0)11(78.6) Menopause6(15.4)3(10.3)3(30.0)3(12.0)3(21.4)Lymph node status0.460.55 Positive22(56.4)15(51.7)7(70.0)15(60.0)7(50.0) Negative17(43.6)14(48.3)3(30.0)10(40.0)7(50.0)Initial clinical stage>0.990.08 Stage II32(82.1)24(82.8)8(80.0)23(92.0)9(64.3) Stage III7(17.9)5(17.2)2(20.0)2(8.0)5(35.7)Ki-67 level>0.99>0.99 Low (≤30%)2(5.1)2(6.9)0(0)1(4.0)1(7.1) High (>30%)37(94.9)27(93.1)10(100)24(96.0)13(92.9)Amount of fibroglandular tissue0.030.67 Fatty or scattered15(38.5)8(27.6)7(70.0)9(36.0)6(42.9) Heterogeneous or extreme24(61.5)21(72.4)3(30.0)16(64.0)8(57.1)Background parenchymal enhancement0.450.35 Minimal or mild26(66.7)18(62.1)8(80.0)18(72.0)8(57.1) Moderate or marked13(33.3)11(37.9)2(20.0)7(28.0)6(42.9)PD-L1 status0.760.005 Positive12(30.8)10(34.5)2(20.0)11(44.0)1(7.1) Negative23(59.0)16(55.2)7(70.0)10(40.0)13(92.9) Unknown4(10.3)3(10.3)1(10.0)4(16.0)0(0)Enhancement pattern>0.99>0.99 Mass29(74.4)21(72.4)8(80.0)19(76.0)10(71.4) Non-Mass10(25.6)8(27.6)2(20.0)6(24.0)4(28.6)Shrinkage Pattern0.400.54 Concentric30(76.9)21(72.4)9(90.0)20(80.0)10(71.4) Non-Concentric9(23.1)8(27.6)1(10.0)5(20.0)4(28.6)Note.—Data are number of patients and data in parentheses are percentages unless otherwise indicated.pCR = pathologic complete response, SD = standard deviation, TFLR = tumor flare-like response. Table 1Characteristics of triple-negative breast cancer patients and imaging features associated with tumor flare-like response (TFLR).Table 1CharacteristicAll (n = 39)TFLR StatuspCR StatusYes(n = 29)No(n = 10)P ValuepCR(n = 25)Non-pCR(n = 14)P ValueAge (years)0.110.51 Mean ± SD45 ± 844 ± 849 ± 845 ± 846 ± 9 Median (range)46(31, 59)44(31,58)45(38,59)44 (31, 59)48(33, 59)Longest diameter of breast tumor(mm)0.170.47 Mean ± SD31.3 ± 10.732.7 ± 10.027.2 ± 12.030.3 ± 10.532.9 ± 11.1 Median (range)29.6(12.3,56.6)32.4(12.3,56.6)22.6(16.7,55.3)29.2(12.3,55.3)31.1(19.8,56.6)Menopausal status0.160.65 Premenopausal33(84.6)26(89.7)7(70.0)22(88.0)11(78.6) Menopause6(15.4)3(10.3)3(30.0)3(12.0)3(21.4)Lymph node status0.460.55 Positive22(56.4)15(51.7)7(70.0)15(60.0)7(50.0) Negative17(43.6)14(48.3)3(30.0)10(40.0)7(50.0)Initial clinical stage>0.990.08 Stage II32(82.1)24(82.8)8(80.0)23(92.0)9(64.3) Stage III7(17.9)5(17.2)2(20.0)2(8.0)5(35.7)Ki-67 level>0.99>0.99 Low (≤30%)2(5.1)2(6.9)0(0)1(4.0)1(7.1) High (>30%)37(94.9)27(93.1)10(100)24(96.0)13(92.9)Amount of fibroglandular tissue0.030.67 Fatty or scattered15(38.5)8(27.6)7(70.0)9(36.0)6(42.9) Heterogeneous or extreme24(61.5)21(72.4)3(30.0)16(64.0)8(57.1)Background parenchymal enhancement0.450.35 Minimal or mild26(66.7)18(62.1)8(80.0)18(72.0)8(57.1) Moderate or marked13(33.3)11(37.9)2(20.0)7(28.0)6(42.9)PD-L1 status0.760.005 Positive12(30.8)10(34.5)2(20.0)11(44.0)1(7.1) Negative23(59.0)16(55.2)7(70.0)10(40.0)13(92.9) Unknown4(10.3)3(10.3)1(10.0)4(16.0)0(0)Enhancement pattern>0.99>0.99 Mass29(74.4)21(72.4)8(80.0)19(76.0)10(71.4) Non-Mass10(25.6)8(27.6)2(20.0)6(24.0)4(28.6)Shrinkage Pattern0.400.54 Concentric30(76.9)21(72.4)9(90.0)20(80.0)10(71.4) Non-Concentric9(23.1)8(27.6)1(10.0)5(20.0)4(28.6)Note.—Data are number of patients and data in parentheses are percentages unless otherwise indicated.pCR = pathologic complete response, SD = standard deviation, TFLR = tumor flare-like response. Characteristics of triple-negative breast cancer patients and imaging features associated with tumor flare-like response (TFLR). Note.—Data are number of patients and data in parentheses are percentages unless otherwise indicated. pCR = pathologic complete response, SD = standard deviation, TFLR = tumor flare-like response. TFLR during NAIC treatment Table 2 summarizes the morphological features of TFLR nodules observed in 29 TNBC patients during NAIC. Median nodule count was 4 (range 1–15). and median largest diameter was 8 mm (range 5–18 mm). Nodules were oval in 27 patients (93.1%) and round in 2 (6.9%), with circumscribed margins and homogeneous internal enhancement. All showed high T2 signal intensity and hyperintensity on diffusion-weighted imaging with facilitated diffusion (non-restrictive) in all cases. Nodules were bilateral in 21 patients (72.5%), ipsilateral in 5 (17.2%), and contralateral in 3 (10.3%) relative to the primary tumor, with consistently asymmetric counts (greater number in one breast than the other). All TFLR nodules were discrete (100%), with no instances observed in clusters. Kinetic curves showed persistent enhancement in 9 patients (31.0%), plateau in 8 (27.6%), and washout in 12 (41.4%).Table 2Morphological features of tumor flare-like response nodules.Table 2Morphological featuresNo.(n = 29)Largest TFLR nodule diameter (mm) Mean ± SD8 ± 3 Median (range)8(5,18)Largest nodule diameter ≥9 mm Yes18 No11Number of TFLR nodules Mean ± SD5 ± 3 Median (range)4(1,15)Shape Round2(6.9) Oval27(93.1) Irregular0(0)Margins Circumscribed29(100) Irregular0(0) Spiculated0(0)Internal enhancement characteristics Homogeneous29(100) Heterogeneous0(0) Rim enhancement0(0) Dark internal septations0(0)Symmetry Yes0(0) No29(100)Location Ipsilateral5(17.2) Contralateral3(10.3) Bilateral21(72.5)Distribution Clusters0(0) Discrete29(100)Kinetic curve pattern Persistent9(31.0) Plateau8(27.6) Washout12(41.4)Note.—Data are number of patients and data in parentheses are percentages unless otherwise indicated. TRLR = tumor flare-like response. Table 2Morphological features of tumor flare-like response nodules.Table 2Morphological featuresNo.(n = 29)Largest TFLR nodule diameter (mm) Mean ± SD8 ± 3 Median (range)8(5,18)Largest nodule diameter ≥9 mm Yes18 No11Number of TFLR nodules Mean ± SD5 ± 3 Median (range)4(1,15)Shape Round2(6.9) Oval27(93.1) Irregular0(0)Margins Circumscribed29(100) Irregular0(0) Spiculated0(0)Internal enhancement characteristics Homogeneous29(100) Heterogeneous0(0) Rim enhancement0(0) Dark internal septations0(0)Symmetry Yes0(0) No29(100)Location Ipsilateral5(17.2) Contralateral3(10.3) Bilateral21(72.5)Distribution Clusters0(0) Discrete29(100)Kinetic curve pattern Persistent9(31.0) Plateau8(27.6) Washout12(41.4)Note.—Data are number of patients and data in parentheses are percentages unless otherwise indicated. TRLR = tumor flare-like response. Morphological features of tumor flare-like response nodules. Note.—Data are number of patients and data in parentheses are percentages unless otherwise indicated. TRLR = tumor flare-like response. TFLR first emerged on MRI after two cycles of NAIC in all 29 patients. Following treatment completion, TFLR completely disappeared in 22 patients (75.9%), while the remaining cases exhibited varying degrees of regression. Kaplan-Meier analysis estimated the median duration as 152 days (95% CI 145–159) (Fig. 2). Median imaging follow-up was 176 days (range 91–244). Seven patients (24.1%) were censored at the last preoperative MRI.Fig. 2Temporal evolution of tumor flare-like response (TFLR) in TNBC patients during preoperative immuno-chemotherapy. Upper Panel: Line graph depicting the temporal changes in the size of the largest TFLR-associated nodule across multiple time points. All 29 patients exhibited the emergence of markedly enhanced nodules on MRI after two cycles of treatment, which subsequently regressed over time. Lower Panel: Kaplan-Meier of TFLR persistence duration (event = complete disappearance of TFLR or last preoperative MRI if persistent, censored). Median duration 152 days. Temporal evolution of tumor flare-like response (TFLR) in TNBC patients during preoperative immuno-chemotherapy. Upper Panel: Line graph depicting the temporal changes in the size of the largest TFLR-associated nodule across multiple time points. All 29 patients exhibited the emergence of markedly enhanced nodules on MRI after two cycles of treatment, which subsequently regressed over time. Lower Panel: Kaplan-Meier of TFLR persistence duration (event = complete disappearance of TFLR or last preoperative MRI if persistent, censored). Median duration 152 days. Histopathological examination of four biopsied TFLR nodules (largest nodule in each patient) revealed a prominent capillary hyperplasia and significant lymphocytes infiltration (Fig. 3). No malignant cells were identified.Fig. 3Representative case of TFLR in a patient with left-sided TNBC. a: Baseline subtracted postcontrast maximum intensitydemonstrates a left breast mass. b: After two cycles of camrelizumab-based immuno-chemotherapy, the mass has markedly decreased while multiple new oval TFLR nodules with circumscribed margins and homogeneous enhancement appear (dashed arrows). The largest measures 9 mm. c: After four cycles, the previously largest nodule has shrunk, but additional TFLR nodules have developed. d: After six cycles (pre-surgery), most TFLR nodules have regressed or disappeared. e, f: Histopathology of the biopsied largest nodule (the 9-mm nodule seen in panel b) shows reactive lymphoid hyperplasia with prominent capillary proliferation (+) and dense lymphocytic infiltration (arrows) (H&E, original magnification ×20). Representative case of TFLR in a patient with left-sided TNBC. a: Baseline subtracted postcontrast maximum intensitydemonstrates a left breast mass. b: After two cycles of camrelizumab-based immuno-chemotherapy, the mass has markedly decreased while multiple new oval TFLR nodules with circumscribed margins and homogeneous enhancement appear (dashed arrows). The largest measures 9 mm. c: After four cycles, the previously largest nodule has shrunk, but additional TFLR nodules have developed. d: After six cycles (pre-surgery), most TFLR nodules have regressed or disappeared. e, f: Histopathology of the biopsied largest nodule (the 9-mm nodule seen in panel b) shows reactive lymphoid hyperplasia with prominent capillary proliferation (+) and dense lymphocytic infiltration (arrows) (H&E, original magnification ×20). Inter-reader agreement Interobserver agreement for TFLR presence/absence was excellent, with Cohen's kappa = 0.92. Maximum diameter measurements of TFLR nodules demonstrated high consistency (ICC = 0.965, 95% CI 0.92–0.98; two-way random-effects, absolute agreement model). Bland-Altman analysis revealed limits of agreement of −1.2 to 1.1 mm (<10% of mean nodule diameter, clinically acceptable), with minor variation mainly in nodules <6 mm (Fig. 4).Fig. 4Bland–Altman plot assessing interobserver agreement for measurement of largest TFLR nodule diameter. Bland–Altman plot assessing interobserver agreement for measurement of largest TFLR nodule diameter. Correlation between TFLR and pathological tumor response Table 3 summarizes the association between TFLR characteristics and pathological tumor response in the study cohort of 39 TNBC patients. Presence of TFLR alone was not significantly associated with pCR (62.1% vs. 70.0%, P = .722). Among the 29 patients with TFLR, median largest nodule diameter was significantly greater in those achieving pCR (9.5 mm vs. 8.0 mm, P = .001, large effect size). Patients with largest nodule ≥9 mm had significantly higher pCR rate (77.8% vs. 36.4%, P = .048).Table 3Association of TFLR with pathological complete response.Table 3CharacteristicpCRNon-pCRPOR (95% CI)TFLR presence(n = 39)0.722 Yes18(62.1)11(37.9)0.70 (0.15-3.18) No7(70.0)3(30.0)RefAmong patients with TFLR(n = 29)Largest nodule diameter(mm)0.001 Mean ± SD10.08 ± 2.767.14 ± 1.92 Median (IQR)9.50(9.12-10.38)8.00 (5.00-9.00)0.597aLargest nodule diameter ≥9 mm0.048 Yes14(77.8)4(22.2)6.13 (1.17-32.10) NO4(36.4)7(63.6)RefNote.—Data are number of patients and data in parentheses are percentages.pCR = pathologic complete response, TFLR = tumor flare-like response.aThe analysis indicated that the effect size for the Mann-Whitney U test is large. Table 3Association of TFLR with pathological complete response.Table 3CharacteristicpCRNon-pCRPOR (95% CI)TFLR presence(n = 39)0.722 Yes18(62.1)11(37.9)0.70 (0.15-3.18) No7(70.0)3(30.0)RefAmong patients with TFLR(n = 29)Largest nodule diameter(mm)0.001 Mean ± SD10.08 ± 2.767.14 ± 1.92 Median (IQR)9.50(9.12-10.38)8.00 (5.00-9.00)0.597aLargest nodule diameter ≥9 mm0.048 Yes14(77.8)4(22.2)6.13 (1.17-32.10) NO4(36.4)7(63.6)RefNote.—Data are number of patients and data in parentheses are percentages.pCR = pathologic complete response, TFLR = tumor flare-like response.aThe analysis indicated that the effect size for the Mann-Whitney U test is large. Association of TFLR with pathological complete response. Note.—Data are number of patients and data in parentheses are percentages. pCR = pathologic complete response, TFLR = tumor flare-like response. The analysis indicated that the effect size for the Mann-Whitney U test is large. Three logistic regression models were evaluated (Table 4, ROC curves in Fig. 5). Univariate analysis of TFLR presence yielded OR = 2.084 (95% CI 1.173–3.701, P = .012) and AUC = 0.861 (95% CI 0.729–0.993). When largest nodule diameter was entered as a continuous variable, the multivariable OR was 2.965 (95% CI 1.298–6.772, P = .01) with AUC = 0.904 (95% CI 0.797–1.00). The multivariable binary model using ≥9 mm cut-off gave OR = 7.833 (95% CI 1.260–48.701, P = .027) and AUC = 0.727 (95% CI 0.519–0.936). All multivariable models adjusted for clinical stage, FGT category, and baseline tumor size; none of the covariates were significant. PD-L1 status was evaluated as a potential confounder but was excluded from the final models due to sparse data, which resulted in unstable estimates. Its inclusion did not materially change the results for largest TFLR nodule diameter.Table 4Logistic regression models for pathological complete response (pCR) in patients with TFLR (n = 29).Table 4ModelOR(95%CI)P- valueAUC(95%CI)SensitivitySpecificityCut-off pointCorresponding SensitivityCorresponding SpecificityPresence model (univariate)2.08(1.17–3.70)0.0120.861(0.729-0.993)83.3%45.5%9.2572.2%90.9%Continuous-size model2.96(1.30-6.77)a0.0100.904(0.797-1)88.9%54.5%0.6977.8%90.9%Binary-size model (≥9 mm vs. <9 mm)7.83(1.26-48.70)a0.0270.727(0.519-0.936)77.8%63.6%0.4994.4%54.5%Note: Cut-off point values probabilities (unitless) or size (mm).aAll models were adjusted for initial clinical stage (stage III VS stage II), FGT (fatty or scattered VS heterogeneous or extreme) and baseline tumor size. PD-L1 status was evaluated as a potential confounder but was excluded from the final models due to sparse data and unstable estimates.Fig. 5ROC curves for pCR among patients with TFLR (n = 29). (A) Univariable model using TFLR presence: AUC = 0.861 (95% CI 0.729–0.993). (B) Multivariable continuous-size model (largest TFLR nodule diameter as continuous variable): AUC = 0.904 (95% CI 0.797–1.00). (C) Multivariable binary-size model (largest TFLR nodule diameter ≥9 mm vs. <9 mm): AUC = 0.727 (95% CI 0.519–0.936). Table 4Logistic regression models for pathological complete response (pCR) in patients with TFLR (n = 29).Table 4ModelOR(95%CI)P- valueAUC(95%CI)SensitivitySpecificityCut-off pointCorresponding SensitivityCorresponding SpecificityPresence model (univariate)2.08(1.17–3.70)0.0120.861(0.729-0.993)83.3%45.5%9.2572.2%90.9%Continuous-size model2.96(1.30-6.77)a0.0100.904(0.797-1)88.9%54.5%0.6977.8%90.9%Binary-size model (≥9 mm vs. <9 mm)7.83(1.26-48.70)a0.0270.727(0.519-0.936)77.8%63.6%0.4994.4%54.5%Note: Cut-off point values probabilities (unitless) or size (mm).aAll models were adjusted for initial clinical stage (stage III VS stage II), FGT (fatty or scattered VS heterogeneous or extreme) and baseline tumor size. PD-L1 status was evaluated as a potential confounder but was excluded from the final models due to sparse data and unstable estimates. Logistic regression models for pathological complete response (pCR) in patients with TFLR (n = 29). Note: Cut-off point values probabilities (unitless) or size (mm). All models were adjusted for initial clinical stage (stage III VS stage II), FGT (fatty or scattered VS heterogeneous or extreme) and baseline tumor size. PD-L1 status was evaluated as a potential confounder but was excluded from the final models due to sparse data and unstable estimates. ROC curves for pCR among patients with TFLR (n = 29). (A) Univariable model using TFLR presence: AUC = 0.861 (95% CI 0.729–0.993). (B) Multivariable continuous-size model (largest TFLR nodule diameter as continuous variable): AUC = 0.904 (95% CI 0.797–1.00). (C) Multivariable binary-size model (largest TFLR nodule diameter ≥9 mm vs. <9 mm): AUC = 0.727 (95% CI 0.519–0.936). Discussion In this study, we identified and characterized tumor flare-like response (TFLR) as a novel imaging phenomenon in TNBC patients undergoing camrelizumab-based NAIC. TFLR, observed in 74.4% of the study cohort, manifested as enhancing nodules separate from the index tumor, appearing uniformly after cycle 2 and regressing by the end of treatment. This is the first study to report such a pattern in TNBC, contributing to the understanding of immune-related imaging changes in breast MRI in the context of immunotherapy. Immune checkpoint inhibitors can induce unconventional response patterns, including pseudoprogression and hyperprogression [[20], [21], [22], [23], [24]]. TFLR resembles tumor flare or pseudoprogression by presenting as “new lesions” [25], yet differs by its benign morphology (oval shape, circumscribed margins, homogeneous enhancement, frequent bilateral distribution in 72.5%) and paradoxical malignant-like kinetics (washout in 41.4%, plateau in 27.6%) [26]. This combination demands careful radiologic interpretation to avoid misdiagnosis. Although small TFLR nodules (<6 mm) may overlap with normal fibroglandular enhancement and showed greater measurement variability on Bland-Altman analysis, overall inter-reader agreement remained high (ICC = 0.965), confirming reproducible identification. All 29 patients developed TFLR after cycle 2, with complete resolution in 75.9% and marked regression in the remainder by treatment completion. Kaplan-Meier analysis estimated the median duration of TFLR persistence to be 152 days (95% CI 145–159). This temporal pattern, opposite to the fibroglandular enhancement suppression typically observed during chemotherapy [27,28], serves as a key diagnostic feature and underscores the importance of recognizing TFLR to prevent misinterpretation as progression. TFLR occurrence was significantly associated with higher FGT density (P = .03). Dense FGT is characterized by greater stromal cellularity and vascularity, which may facilitate immune cell infiltration and create a microenvironment conducive to TFLR development [[29], [30], [31]]. No association was observed with BPE, likely due to BPE's greater hormonal and cyclic variability. Histopathological findings of four biopsied nodules revealed prominent capillary hyperplasia and significant lymphocytic infiltration, suggesting an immune-mediated etiology. Camrelizumab's known VEGF-stimulating effect offers a plausible mechanism for the observed angiogenesis [[32], [33], [34], [35]]. Interestingly, no significant increase in PD-L1 expression was observed in resected tumors from TFLR patients, which suggests that off-tumor immune activation may be distinct from direct anti-tumor immunity. Further studies are needed to explore the role of immune cell subsets, cytokines, and the tumor microenvironment in TFLR development. Overall, the presence of TFLR was not associated with pCR (P = .72). The ≥5 mm detection threshold excluded smaller immune-related nodules, a lower threshold might have revealed an association by capturing milder immune activation. ROC analysis with bootstrap validation identified 9 mm as the optimal cut-off for the largest TFLR nodule, likely marking the threshold at which off-tumor immune activation becomes therapeutically effective. In contrast, larger nodule size, reflecting more intense off-tumor immune activation, was a strong, independent predictor of pCR in multivariable analysis (continuous variable: OR = 2.96, 95% CI 1.30–6.77, P = .01; ≥9 mm vs. <9 mm: OR = 7.83, 95% CI 1.26–48.70, P = .027). This graded relationship suggests that only a robust stromal immune response, as manifested by bigger TFLR nodules, parallels clinically meaningful anti-tumor immunity. This study has several limitations. The small sample size and single-arm, single-center design limit generalizability, increase the risk of type II error and overfitting, and were not powered for this imaging sub-analysis. Findings are specific to camrelizumab-based immuno-chemotherapy and require validation in larger, multi-center cohorts with diverse regimens. Histologic correlation was restricted to four non-systematically sampled nodules; immunohistochemistry and serial biopsies were not performed, precluding detailed mechanistic insight. The ≥5 mm detection threshold excluded smaller immune-related nodules, potentially underestimating milder forms of immune activation. Follow-up was short, preventing assessment of long-term prognostic implications. Although readers were blinded to pCR outcome and underwent consensus training on de-identified cases, retrospective identification of a novel phenomenon not described in initial clinical reports raises the possibility of recognition bias. These limitations highlight the need for confirmatory studies in independent cohorts. Conclusions This study identifies Tumor flare-like response as a novel imaging phenomenon in TNBC patients undergoing camrelizumab-based NAIC. Largest TFLR nodule diameter ≥9 mm is a strong, independent predictor of pathological complete response and represents a promising early, non-invasive imaging biomarker of immunotherapeutic efficacy. Recognizing TFLR is essential to avoid misinterpretation as disease progression and to guide optimal management of TNBC patients treated with immunotherapy. Prospective validation in larger, multi-center cohorts is warranted. CRediT authorship contribution statement Bing Jie Zheng: Writing – review & editing, Writing – original draft. Cheng Zheng Wang: Supervision, Data curation. Ho-Young Song: Writing – review & editing, Supervision. Wen Juan Xu: Validation, Data curation. Ren Zhi Zhang: Methodology. Jin Rong Qu: Validation, Methodology. Chun Miao Xu: Validation. Xia Yu Yue: Resources, Data curation. Jia Yu Wang: Methodology, Investigation. Yu Chen: Methodology, Investigation. Ming Ge Liu: Methodology, Investigation. Cong Yu Li: Validation, Investigation. Ya Fei Chen: Visualization, Software, Formal analysis. Quan Li Gao: Validation, Supervision. Xue Jun Chen: Supervision, Methodology. Xiang Li: Validation, Supervision. Zhen Zhen Liu: Funding acquisition, Conceptualization. Ethics approval and consent to participate This study was approved by the Institutional Review Board of Henan Cancer Hospital (approval number: 2019110712). Written informed consent was obtained from all participants prior to enrollment. Consent for publication All participants provided consent for the publication of anonymized data. Availability of data and materials The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. Funding This study was funded by the Joint Co-construction Project of Henan Provincial Medical Science and Technology Research Program (No.LHGJ20230107), National High Level Hospital Clinical Research Funding and National Cancer Center Climbing Fund (No. NCC202416005). Conflict of interest statement All authors disclose no financial and personal relationships with other people or organisations that could inappropriately influence (bias) our work.