A Randomized, Phase II Clinical Trial of FLT-PET and FDG-PET for Early Response Assessment of Neoadjuvant Systemic Therapy in Triple-Negative Breast Cancer Authors: Melek Akay, Jennifer Glendenning, Holly Tovey, Yi-ting Tsai, Rosalind Graham, Esme Carpenter, Helen Kakkassery, Thanussuyah Alaguthurai, Laura Finneran, Ioannis Roxanis, Angela Swampillai, Mark Harries, Ines Sandri, Ravindran Karthigan, Sally Barrington, Patrycja Gazinksa, Gary Cook, Sugama Chicklore, Judith Bliss, Syed Haider, Andrew Tutt, Sheeba Irshad Published: 2026 ABSTRACT AbstractPurpose:Early identification of response to neoadjuvant chemotherapy (NACT) in triple-negative breast cancer (TNBC) can facilitate timely treatment adjustments. This phase II analytic and clinical validity study (TNPET01) evaluated whether [18F]-fluorodeoxyglucose (FDG) or [18F]-fluorothymidine (FLT) positron emission tomography (PET)/CT can predict response after one NACT cyclePatients and Methods:In part A (analytic validity phase), patients with stage II–III TNBC were randomized to FDG or FLT imaging. Baseline repeat scans assessed test–retest repeatability, followed by a post–cycle 1 scan in week 3. Dynamic imaging preceded static acquisitions at 90-, 120-, and 180-minute (FDG) or 90-minute (FLT), evaluating SUVmax, SUVmean, SUVpeak, and SULpeak. Tracer selection for part B was based on prespecified repeatability and response criteria. Part B (clinical validity phase) examined associations between changes in SUV (ΔSUV) after one cycle and post–cycle 3 MRI, end-of-treatment MRI, and residual cancer burden (RCB) at surgery. Exploratory analyses assessed relationships between PET response, Ki-67, and tumor-infiltrating lymphocytes (TIL).Results:Twenty-two patients were enrolled. Both tracers met repeatability thresholds; FDG was selected for part B owing to superior image quality and availability. Fourteen patients underwent FDG-PET across both parts. Reductions in SUVmax/mean after one cycle significantly correlated with mid-treatment MRI response and final RCB score (P < 0.005). Early post–cycle 1 increases in TILs correlated with greater metabolic reduction and lower RCB, whereas Ki-67 changes were not predictive.Conclusions:FDG-PET after one NACT cycle was associated with histologic response and stronger correlations with RCB than MRI. These findings support PET as an early biomarker for treatment response and warrant validation in larger, immunotherapy-era trials. Translational RelevancePatients with triple-negative breast cancer (TNBC) and extensive residual disease after neoadjuvant chemotherapy (NACT) face a significantly worse prognosis than those with favorable pathologic response. Our study demonstrates that an early [18F]-fluorodeoxyglucose (FDG)–positron emission tomography (PET) scan after one NACT cycle is significantly associated with subsequent pathologic response and shows stronger associations with residual cancer burden than RECIST-based midtreatment MRI within this cohort. This has significant clinical implications; early identification of unfavorable responders could facilitate timely treatment escalation, whereas strong metabolic responders may be eligible for treatment de-escalation. Additionally, we demonstrate a link between early metabolic response and chemotherapy-induced immune activation, with increases in tumor-infiltrating lymphocytes correlating strongly with both post–cycle 1 metabolic reduction and subsequent pathologic response. In the modern chemoimmunotherapy era, this post–cycle 1 window may capture true tumor response before the later onset of confounding immune effects like pseudoprogression. These findings strongly support early FDG-PET as a robust biomarker to guide adaptive therapy in neoadjuvant TNBC treatment and warrant validation in larger, immunotherapy-containing trials. Translational Relevance Patients with triple-negative breast cancer (TNBC) and extensive residual disease after neoadjuvant chemotherapy (NACT) face a significantly worse prognosis than those with favorable pathologic response. Our study demonstrates that an early [18F]-fluorodeoxyglucose (FDG)–positron emission tomography (PET) scan after one NACT cycle is significantly associated with subsequent pathologic response and shows stronger associations with residual cancer burden than RECIST-based midtreatment MRI within this cohort. This has significant clinical implications; early identification of unfavorable responders could facilitate timely treatment escalation, whereas strong metabolic responders may be eligible for treatment de-escalation. Additionally, we demonstrate a link between early metabolic response and chemotherapy-induced immune activation, with increases in tumor-infiltrating lymphocytes correlating strongly with both post–cycle 1 metabolic reduction and subsequent pathologic response. In the modern chemoimmunotherapy era, this post–cycle 1 window may capture true tumor response before the later onset of confounding immune effects like pseudoprogression. These findings strongly support early FDG-PET as a robust biomarker to guide adaptive therapy in neoadjuvant TNBC treatment and warrant validation in larger, immunotherapy-containing trials. Introduction Triple-negative breast cancer (TNBC) accounts for 10% to 15% of all breast cancer diagnoses and is characterized by the absence of expression of the estrogen and progesterone receptors and lack of human epidermal growth factor receptor 2 (HER2) overexpression (1). The mainstay of treatment for stage II–III TNBC is neoadjuvant systemic therapy to downstage tumors, followed by definitive surgery. Residual cancer burden (RCB) after neoadjuvant chemotherapy (NACT) has been shown to be prognostic for long-term survival, with patients who achieve a pathologic complete response (pCR), i.e., complete absence of residual invasive disease, or minimal RCB (RCB-I), observed to achieve good prognoses (2), whereas increased extent of residual disease (RCB-II or RCB-III) has been associated with higher risk of recurrence (3). Despite the approval of immunotherapy with pembrolizumab to chemotherapy (4), approximately 20% of patients continue to exhibit chemotherapy- and immunotherapy-resistant disease (5), associated with rapid incurable relapse, limited treatment options, and short survival. Early identification of patients unlikely to benefit from standard neoadjuvant therapy is crucial. Although breast MRI is considered the optimal tool for post-NACT assessment of residual disease (6, 7), its reliability for on-treatment response monitoring is less clear. Constraints include the lack of standardized response criteria beyond anatomic size reduction (8) and the inherent delay in measurable tumor shrinkage, which typically requires at least three treatment cycles. In contrast, 2-[fluorine18]fluoro-2-deoxy-D-glucose (FDG)–positron emission tomography (PET) can detect metabolic changes in tumors before morphologic shrinkage and is increasingly being explored as an early response biomarker (9, 10). The phase II PHERGain trial demonstrated the feasibility of PET-guided response adaptation in HER2-positive breast cancer (11), but a subsequent analysis revealed that the optimal SUVmax reduction threshold for predicting pCR was considerably higher than initially used (>77%; ref. 12). More broadly, there is no consensus on robust metabolic response thresholds. The 2024 joint European Association of Nuclear Medicine (EANM)–Society of Nuclear Medicine and Molecular Imaging (SNMMI) guidelines (9) recommend using established criteria such as those of the European Organisation for Research and Treatment of Cancer (EORTC) or PET Response Criteria in Solid Tumors (PERCIST). However EORTC criteria (13) defined PET response as a 15% SUVmax reduction after 1 cycle, whereas PERCIST guidelines (14) advocate a stricter 30% SUVpeak reduction, and more recent early-stage breast cancer studies suggest thresholds of 40% to 77% (12, 15–17). This challenge is likely amplified in TNBC in which tumors generally exhibit lower and more heterogeneous baseline FDG avidity compared with HER2-positive disease (18). Although TNBC-specific studies have demonstrated correlations between early changes in SUVmax and histopathologic response following NACT (15, 16, 19), PET-based response assessment in this setting is not yet routine (20). Uncertainty remains regarding tracer selection, optimal standardized uptake value (SUV) parameter, and acquisition timing. Among these uncertainties is tracer selection. Although FDG tracer is the standard oncology tracer, its uptake reflects glucose metabolism and may accumulate in inflammatory tissue. An alternative tracer, 3′deoxy-3′-[18F]fluorothymidine (FLT) reflects DNA synthesis and may offer greater specificity, but comparative clinical data in TNBC are limited (21, 22). Uncertainty also persists regarding the most appropriate SUV parameter and imaging time point. SUVmax, reflecting the single hottest voxel in the tumor, is widely used but is susceptible to imaging noise and partial volume effects (23). SUVmean averages uptake across a defined tumor volume, reducing noise but is highly dependent on choice of volume and tumor heterogeneity (24). PERCIST guidelines recommend SUVpeak, which averages uptake in the hottest 1 cm3 of the tumor, with lean body mass correction for consistency (SULpeak). However the optimal metric for TNBC remains undefined. Similarly, although imaging is conventionally performed 60 to 90 minutes after tracer injection, delayed imaging up to 180 minutes may improve lesion detectability (25, 26). Here, we aimed to address these uncertainties through TNPET01, a feasibility and response prediction study in TNBC, designed to evaluate baseline and early on-treatment FDG-PET changes and their correlation with MRI, biopsy, and pathologic response to NACT. Patients and Methods Study design TNPET01 was a randomized, open-label, phase II clinical trial. The two-part study evaluated the repeatability and response prediction of measurements of two PET tracers, FDG and FLT (part A), and then used the optimal tracer to correlate early metabolic response after one NACT cycle with radiologic and pathologic outcomes (part B; Fig. 1A). The study was approved by the Research Ethics Committee (REC number: 11/L0/1492) and the Medicines and Healthcare products Regulatory Agency. All procedures involving human participants were conducted in accordance with the ethical standards of the institutional research committee and with the Declaration of Helsinki and its later amendments. All patients provided written informed consent. Study design and patient flow. A, Schematic overview of the TNPET01 study. Part A evaluated the test–retest repeatability of FDG and FLT PET tracers in early-stage TNBC to inform tracer selection. Part B evaluated early metabolic response using FDG-PET after one cycle of NACT and its association with downstream radiologic and pathologic outcomes. B, CONSORT-style flow diagram of patient enrollment, randomization, tracer allocation, imaging time points, and evaluability across parts A and B. In part A, 2 patients were not evaluable owing to withdrawal of consent (n = 1) and FLT tracer production failure (n = 1). In part B, 2 patients were not evaluable owing to withdrawal of consent (n = 1) and ineligibility confirmed after randomization (n = 1). C1, cycle 1; EOT, end of treatment. [Created in BioRender. Akay, M. (2026) https://biorender.com/i4ud4mk]. Eligibility criteria Eligible patients were women >18 years with stage II–III estrogen receptor–negative (Allred <3) and HER2-negative (IHC 0/1+ or IHC2+ and nonamplified FISH) invasive breast cancer who were planned to start NACT. Exclusion criteria included pregnancy, breastfeeding, or diabetes mellitus. Treatment protocol TNPET01 was conducted prior to the adoption of the anti–PD-1 antibody pembrolizumab into NACT protocols. NACT consisted of four cycles of epirubicin 90 mg/m2 with cyclophosphamide 600 mg/m2 (EC) every 3 weeks and four cycles of docetaxel (T) 100 mg/m2 every 3 weeks. Imaging protocol In part A, patients were randomized at a 1:1 ratio to receive the FDG or FLT tracer. Each patient underwent two pretreatment PET/CT scans, separated by 1 to 5 days, to assess uptake repeatability. Static scan acquisitions were performed at 90, 120, and 180 minutes after tracer injection in the FDG group and at 90 minutes after injection in the FLT group. A further PET scan with the same tracer was then performed after cycle 1 NACT at day 17 ± 3. For each PET time point, tumor volumes of interest (VOI) were generated using a 40% threshold (27) of the maximum uptake within a manually defined area of interest. SUVmax (the hottest voxel within the 40% isocontour), SUVmean (the average SUV across the VOI), SUVpeak (the mean SUV within the hottest 1 cm3 volume of the tumor), and SULpeak (SUVpeak corrected for lean body mass) were recorded. Given the log-normal distribution of SUV, measurements were log-transformed prior to part A analyses to improve statistical robustness. The optimal tracer for part B was selected from part A following analysis of tracer repeatability, according to predefined criteria below. In part B, patients had one baseline and one post–cycle 1 (day 17 ± 3) PET/CT scan. Patients in both parts had a midtreatment MRI following three NACT cycles and an end-of-treatment (EOT) MRI as per standard practice. In part B, an optional post–cycle 1 (day 17 ± 3) diffusion-weighted (DW) MRI was included to assess early changes in the apparent diffusion coefficient (ADC). PET/CT and MRI acquisition protocols remained unchanged throughout the trial. PET was performed for research only and was not used for clinical decision-making. All PET scans were analyzed locally by a single experienced nuclear medicine physician using a predefined workflow, with semiautomated threshold-based VOI generation. For response analyses, a single index lesion per patient was analyzed. When multiple lesions were present, the lesion demonstrating the highest FDG uptake on baseline PET (“hottest lesion”) was selected and tracked across subsequent scans. This lesion was typically the primary breast tumor but could include nodal metastases where uptake predominated. Primary and secondary analyses Analyses were conducted according to the predefined study parts and corresponding analysis populations described below. The part A primary analysis was repeatability of FDG and FLT tracers, defined as at least one SUV/SUL parameter demonstrating lack of variability within ±15%. For a tracer to advance to part B, a 20% to 40% reduction in SUV/SUL was required in ≥50% of responders on post–cycle 3 MRI (the predefined imaging comparator). This range was selected a priori as a pragmatic feasibility criterion, informed by published early metabolic response thresholds (14, 28), and was intended to identify tracers demonstrating biologically relevant signal change while accounting for measurement variability in small imaging cohorts. It was not intended to define a definitive predictive cutoff. Safety of the FLT tracer was also evaluated. In part B, the primary analysis assessed correlation between post–cycle 1 ΔSUV and RECIST-defined tumor size change on post–cycle 3 MRI to determine whether early metabolic change predicted later radiologic response. Secondary analyses assessed correlation between ΔSUV and pathologic outcome using the RCB score. Additional secondary analyses included correlation between ΔSUV and EOT MRI and evaluation of the optimal PET acquisition time point. Among prespecified acquisition time points (90, 120, and 180 minutes), the optimal time point was determined based on predefined analytic criteria, including repeatability (part A), evaluability across lesions, and consistency of response separation. Two analysis populations were defined following a per-protocol approach: (i) the part A analytic validation cohort, used to assess tracer repeatability and inform parameter selection, including all patients in part A who received both baseline scans for the relevant tracer, and (ii) the pooled FDG cohort (part A + part B), used for correlation analyses following tracer selection to maximize evaluable FDG data consistent with the study design, including all patients treated with the tracer which progressed to part B who had the assessments relevant for each analysis, regardless of whether or not they completed chemotherapy as planned. Pathology assessment Pathologic response was assessed at surgery using the RCB classification. The RCB index incorporates dimensions of residual primary tumor, cellularity, and nodal burden to generate a continuous score and categorical class (RCB-0, I, II, and III). RCB scores were calculated using the MD Anderson Online Calculator (https://www3.mdanderson.org/app/medcalc/index.cfm?pagename=jsconvert3). In this study, RCB-0 [pCR ± residual ductal carcinoma in situ (DCIS)] and RCB-I were grouped to define favorable responders, whereas RCB-II and RCB-III were considered unfavorable responders (29). This grouping serves a pragmatic analytic purpose and does not signify identical long-term outcomes for the subgroups within each combined group. Research core biopsies were obtained at baseline, after cycle 1, and surgery for exploratory biomarker analyses. All IHC evaluations were performed centrally by a dedicated breast pathologist. In confirmed carcinoma cases [using hematoxylin and eosin (H&E)–stained sections], stromal tumor-infiltrating lymphocytes (TIL) were quantified as per Salgado and colleagues (30). Ki-67 expression was assessed on formalin-fixed, paraffin-embedded sections (antibody clone 30-9, RRID: AB_2631262) and scored as the percentage of positively stained tumor cell nuclei (31). Biomarker analyses required the presence of evaluable invasive carcinoma. Post–cycle 1 biopsies lacking residual carcinoma or containing insufficient tumor cells were therefore considered nonevaluable for Ki-67 and TIL assessment. Statistical considerations The study aimed to evaluate the correlation between ΔSUV for the part B tracer with RECIST-defined tumor size percentage change on post–cycle 3 MRI. To establish a correlation of ≥85% while excluding correlations <60%, a sample size of 20 subjects was required in part B (α = 10%, one-sided; power = 85%). The primary efficacy analysis was performed on a per-protocol population, which included all participants who completed the required imaging assessments for part B. Attrition and protocol deviations were monitored throughout the study. In part A, a sample size of 5 patients per tracer was selected based on comparable data in other tumor sites to demonstrate repeatability and was not powered for direct comparison between tracers. Randomization to tracer was used to optimize the balance of tumor and patient factors that might affect performance assessment, was stratified by nodal involvement (yes/no), and conducted centrally by the Clinical Trials and Statistics Unit at the Institute of Cancer Research. Repeatability was assessed at both patient and lesion-by-time point level. Patient-level analysis summarized each SUV parameter using the average across all lesions and scan times. Repeatability coefficients were calculated according to Bland and Altman methods and visualized in Bland–Altman plots (32). For the lesion-by-time point analyses, a mixed-effects model accounted for the nested data structure. For part B, Spearman’s rank correlation assessed the relationship between ΔSUV and MRI-based tumor size reduction (primary endpoint). If multiple lesions were present, the hottest lesion at baseline was selected for response monitoring. Analyses of secondary endpoints are descriptive. All statistical analyses were performed using Stata (RRID: SCR_012763) v11 (part A) and Stata v17 (part B), with data from part A contributing to part B endpoint analyses. Results Patient characteristics Twenty-two patients consented across parts A and B (Fig. 1A and B) between October 2012 and February 2017. In part A, 11 patients were randomized: 6 to the FLT group and 5 to the FDG group (Fig. 1B). All 5 patients in the FDG group were included in the final analysis, whereas in the FLT group, one patient withdrew consent after the first scan and another was withdrawn because of tracer production failure, resulting in 4 patients in the final analysis. For part B, 11 additional patients were recruited into the FDG group. After one withdrawal and one ineligibility (confirmation of HER2 positivity after randomization), 14 patients were evaluable for part B analyses (5 from part A and 9 from part B; Fig. 1B). Baseline patient and tumor characteristics by tracer group are summarized in Table 1. Of the 14 patients, 2 did not complete the planned eight cycles of chemotherapy: one progressed after three cycles and had RCB-II disease at surgery, and one discontinued after five cycles because of toxicity but achieved pCR (RCB-0). One patient in the FLT group switched early to EC because of progression on docetaxel and subsequently achieved pCR. All patients with the relevant assessments were included in the corresponding analyses. The apparent imbalance in chemotherapy sequencing between FLT and FDG cohorts reflects an institutional change in standard TNBC practice during the study, from taxane followed by EC (T-EC) to EC followed by a taxane (EC-T), and was unrelated to tracer allocation or PET analysis. Table 1.Clinicopathologic characteristics of the study cohort according to tracer group.CharacteristicFLT (n = 4)FDG (n = 14)aOverall (n = 18)Age (median, range)39 (30–53)47 (36–52)46 (30–53)Race (n, %)​​​ White3 (75)8 (57.1)11 (61.1) Black1 (25)4 (28.6)5 (27.8) Asian0 (0)2 (14.3)2 (11.1)BMI (median, range)23.5 (18–29)27.2 (20–32)26 (18–32)Grade (n, %)​​​ 21 (25)2 (14.3)3 (16.7) 33 (75)12 (85.7)15 (83.3)Tumor (mm)​​​ Median, range31 (25–38)27.5 (13–55)28 (13–55)Nodal status (n, %)​​ Node-positive1 (25)6 (42.9)7 (38.9) Node-negative3 (75)8 (57.1)11 (61.1)ER status (n, %)​​​ 04 (100)12 (85.7)15 (83.3) 1000 202 (14.3)3 (16.7)HER2 status (n, %)​​​ 0–14 (100)13 (92.9)17 (94.4) FISH −ve01 (7.1)1 (5.6)NACT (n, %)​​​ T-EC4 (100)9 (64.3)13 (72.2) EC-T05 (35.7)5 (27.8)Primary tumor size is according to MRI longest dimensions (mm).Abbreviations: BMI, body mass index; ER, estrogen receptor.aFDG cohort includes patients from part A and part B following tracer selection for early response analyses. Clinicopathologic characteristics of the study cohort according to tracer group. Primary tumor size is according to MRI longest dimensions (mm). Abbreviations: BMI, body mass index; ER, estrogen receptor. FDG cohort includes patients from part A and part B following tracer selection for early response analyses. Tracer repeatability Part A aimed to identify the optimal tracer for early response assessment in TNBC by evaluating repeatability of baseline SUVs. Patients underwent two pretreatment scans with either FDG or FLT tracer, separated by 3 to 5 days. All images were deemed evaluable by an experienced nuclear medicine physician. Breast lesions were visualized in all patients, and axillary lesions identified in 3 patients (1 FLT and 2 FDG), yielding five evaluable lesions for the FLT group and seven for FDG (Fig. 2A and B). Comparative repeatability of FDG and FLT tracers and early predictive performance in TNBC. A, Representative [18F]FDG-PET/CT images acquired 90 minutes after injection showing tracer uptake in a primary breast tumor (left) and a corresponding axillary lymph node (right). Upper panels show CT-only slices; lower panels show PET/CT fusion. B, Representative [18F]FLT-PET/CT images at 90 minutes after injection from a different patient, demonstrating uptake in the primary breast tumor (left) and axillary lymph node (right). C, Bland–Altman plots of log-transformed SUV values from repeat baseline scans at the 90-minute time point, assessing the test–retest variability for: SUVmax, SUVmean, SUVpeak, and SULpeak. Each point represents a single lesion. Red and blue dots correspond to FDG and FLT tracer groups, respectively. The gray shaded region represents a 15% variability threshold (log ± 0.14). D, Representative [18F]FDG-PET/CT and MRI images from a patient demonstrating marked treatment response. Top and middle, baseline and post–cycle 1 FDG-PET/CT images showing a marked reduction in tumor metabolic activity. Bottom, baseline and post–cycle 3 contrast-enhanced MRI (post-contrast subtracted) showing corresponding volumetric tumor shrinkage. E, Individual patient-level FDG-PET responses following one cycle of NACT, shown for ΔSUVmax (top) and ΔSUVmean (bottom) at 90, 120, and 180 minutes after injection. Bars are colored according to post–cycle 3 MRI response [complete response (CR), partial response (PR), stable disease (SD), or progressive disease (PD)]; patterned bars indicate inclusion of evaluable nodal lesions. F, Individual patient-level FLT-PET responses after one cycle of NACT, shown for ΔSUVmax (left) and ΔSUVmean (right) at 90 minutes after injection. Bars are annotated with patient study ID and colored by post–cycle 3 MRI response category. *, Patient 7 underwent an earlier MRI after two cycles because of clinical signs of progression. C1, cycle 1; C2, cycle 2; C3, cycle 3. Tracer repeatability was assessed using Bland–Altman analyses across all static imaging time points (Fig. 2C; Supplementary Fig. S1A and S1B). SUVmax and SUVmean were measurable across all lesions, whereas SUVpeak could not be obtained for one FLT axillary lesion and three FDG lesions (Supplementary Tables S1 and S2). In a lesion-by-time point sensitivity analysis at 90 minutes, FDG demonstrated repeatability within 12% for SUVpeak and SULpeak, increasing to 24% for SUVmax (Supplementary Table S3). Greater fluctuation was observed at delayed time points, with SUVmean variability reaching 28% at 120 minutes (Supplementary Fig. S1A; Supplementary Table S3). FLT showed similar overall variability at 90 minutes, with SUVmax fluctuating by up to 23% (Fig. 2C; Supplementary Table S3). Patient-level repeatability coefficients (Supplementary Table S4) report the percentage change between scans for each SUV parameter. For patients with multiple lesions, SUV parameters were averaged across lesions and time points. FDG displayed superior consistency, with repeatability within ±15% for SUVmean and SUVmax and within ±16% for SUVpeak and SULpeak. Although FLT matched FDG for SUVpeak and SULpeak, variability was higher for SUVmean (±20%) and SUVmax (±25%). A representative case demonstrating marked metabolic and radiologic response is shown in Fig. 2D. Having confirmed tracer repeatability, predictive performance following one cycle of NACT was assessed. Percentage changes for SUV parameters between the second baseline scan and the post–cycle 1 scan were calculated. All patients who achieved partial or complete radiologic response on post–cycle 3 MRI showed ≥20% reduction in all SUV parameters on post–cycle 1 PET, meeting the predefined response criteria (Fig. 2E and F; Supplementary Fig. S1C and S1D). In the FDG group, all patients exceeded the PERCIST-defined 30% response threshold for SUV/SULpeak reduction (Supplementary Fig. S1C). In the FLT group, one patient with clinical progression underwent an early MRI after two cycles of docetaxel, confirming radiologic progression and showed an increase in SUVmax and SUVpeak, along with a <5% reduction in SUVmean and SULpeak (Fig. 2F; Supplementary Fig. S1D). In summary, both FDG and FLT tracers met the predefined repeatability and response criteria. Due to superior consistency across multiple SUV parameters and broader practical availability, FDG was selected for part B. Association of early FDG-PET response with radiologic and pathologic outcomes FDG patients from part A and part B were pooled for part B analyses following tracer selection, giving a total of 14 evaluable patients. The prespecified primary endpoint for part B was the correlation between ΔSUV after cycle 1 and post–cycle 3 MRI response. Based on per-patient repeatability performance in part A and consistent evaluability across all lesions, SUVmax and SUVmean were selected over peak values for response assessment. In patients with multiple lesions (n = 8), the hottest baseline lesion was selected for monitoring; breast in five cases and axillary nodes in three. Both ΔSUVmax and ΔSUVmean at cycle 1 showed positive correlations with post–cycle 3 MRI tumor reduction, with significance reached at 90 (r = 0.61; P = 0.047) and 120 minutes (r = 0.68; P = 0.015) for SUVmax and similar correlations for SUVmean (Fig. 3A). These findings fulfill the prespecified primary endpoint of part B, demonstrating that early metabolic change after one cycle is associated with subsequent radiologic response. Association between ΔSUVmax/mean after one cycle and subsequent response parameters. A, Spearman correlation matrix showing associations between percentage change in SUVmax and SUVmean after one cycle of NACT (ΔSUV) at 90, 120, and 180 minutes and four response-related endpoints: change in tumor size on post–cycle 3 MRI, change in ADC on post–cycle 1 MRI, change in tumor size on EOT MRI, and the continuous RCB index. For ADC, correlations were based on evaluable breast tumors only. For all other endpoints, the “hottest” lesion per patient (breast or axilla) was used. Each circle represents a correlation, with color indicating strength/direction and size reflecting magnitude. Numerical P values are overlaid. B, Percentage change in SUVmax and SUVmean after one cycle of NACT, shown at 90, 120 and 180 minutes after injection. Bars represent the median ΔSUV values within each pathologic response group: responders (RCB-0/I, shown in blue) and nonresponders (RCB-II/III, brown). Statistical comparison between the two response groups was performed using the Mann–Whitney test with false discovery rate correction (adjusted P < 0.005 across all time points for both parameters). C and D, Serial MRI-based quantification of tumor size at baseline, after cycle 1, and after cycle 3 in (C) a patient who achieved a pCR (RCB-0) and (D) a patient with extensive RCB-III. Post-contrast subtracted images are shown to enhance tumor conspicuity for MRI images. E, Percentage change in ADC values (mm2/s) between baseline and after cycle 1 for patients with RCB-0/I (n = 2) versus RCB-II/III (n = 3). No significant difference was observed in ADC percentage change between the two response groups. F, Association between RECIST-defined MRI response and subsequent RCB group. C1, cycle 1; C3, cycle 3; CR, complete response; ns, nonsignificant; PD, progressive disease; PR, partial response; SD, stable disease. To further validate ΔSUV as a predictive marker, we examined its correlation with other imaging endpoints and response markers (Fig. 3A). Optional post–cycle 1 DW-MRI was available for 5 patients. No significant correlation was observed between ΔSUVmax/mean and ADC change following cycle 1 or EOT (Fig. 3A). As an exploratory analysis, we next assessed correlation with RCB, a clinically validated pathologic endpoint in neoadjuvant TNBC. Looking first at the continuous RCB index, a prespecified pathologic endpoint, ΔSUVmax and ΔSUVmean after one cycle showed strong correlations across all time points (P < 0.005; Fig. 3A). For descriptive comparison, patients were then classified into favorable versus unfavorable response groups (RCB-0/I vs. RCB-II/III). Of the 14 patients in the FDG group, 8 (57%) achieved RCB-0/I, whereas 6 (43%) had RCB-II/III disease. Baseline clinical characteristics were comparable between the two groups, with no features predictive of response (Table 2). Similarly, baseline PET metabolic activity was not associated with pathologic response in breast tumors (P > 0.99 for SUVmax at 90 minutes and P = 0.85 for SUVmean at 90 minutes) or axillary nodes (P = 0.57; Table 2). Table 2.Clinical, histologic, and PET parameters according to RCB score among those who received FDG tracer.Clinical and PET parametersRCB-0/RCB-I (n = 8)RCB-II/III (n = 6) P valueMedian age (years)45.547.50.54Median tumor size (mm)27.528.50.83 T-score (n)​​​ T11 (12.5)1 (16.7)>0.99 T26 (75)4 (66.7)​ T31 (12.5)1 (16.7)​Grade (n)​​​ 21 (12.5)1 (16.7)0.69 37 (87.5)5 (83.3)​Nodal involvement​​​ Positive3 (37.5)3 (50)0.53 Negative5 (62.5)3 (50)​Baseline FDG SUVmax, 90 minutes​​​ Breast lesions (n)86​ Median (range)13.79 (5.85–19.55)12.01 (7.01–25.85)>0.99 Nodal lesions (n)53​ Median (range)4.64 (2.71–15.95)4.9 (3.59–13.33)0.57Baseline FDG SUVmean 90 minutes​​​ Breast lesions (n)86​ Median (range)8.44 (3.36–11.81)7.58 (4.27–16.81)0.85 Nodal lesions (n)53​ Median (range)2.83 (1.59–7.01)2.88 (2.13–8.47)0.57Baseline PET parameters are from the second baseline scan in part A and the single baseline scan in part B. Fisher exact test used to compare proportional values and Mann–whitney test for comparing median values within the two response groups. Clinical, histologic, and PET parameters according to RCB score among those who received FDG tracer. Baseline PET parameters are from the second baseline scan in part A and the single baseline scan in part B. Fisher exact test used to compare proportional values and Mann–whitney test for comparing median values within the two response groups. We then examined whether the association between early metabolic change and subsequent RCB index was maintained when analyzing the categorical response groups. Comparing ΔSUVmax and ΔSUVmean between the two response groups revealed statistically significant differences across all time points [false discovery rate (FDR)-adjusted P < 0.005 at 90, 120, and 180 minutes; Fig. 3B]. Among favorable responders, the percentage reduction in SUVmax and SUVmean ranged from 46.9% (SUVmax, 90 minutes) to 86.9% (SUVmean, 180 minutes), whereas in unfavorable responders, the reduction ranged from 8.3% (SUVmax, 90 minutes) to 47.3% (SUVmean, 90 minutes). Representative MRI examples of patients achieving pCR and extensive residual disease are shown in Fig. 3C and D. Finally, in the 5 patients for whom post–cycle 1 ADC data were available, early changes in tumor ADC did not differentiate favorable (RCB-0/I) from unfavorable responders (RCB-II/III; Fig. 3E; Supplementary Table S5). To contextualize this FDG-PET performance, we evaluated the predictive value of MRI for pathologic outcome. Among 17 patients with evaluable post–cycle 3 MRI (across both tracers), RECIST-defined radiologic responses (CR/PR vs. SD/PD) were not significantly associated with the RCB score at surgery (P = 0.153, Fisher exact test). Similarly, EOT MRI response was not predictive of pathologic outcome (n = 16; P = 0.063; Fig. 3F). Together, these findings suggest that neither RECIST-based tumor shrinkage nor early ADC change were significantly associated with pathologic response in this small cohort. In contrast, FDG-PET–derived ΔSUVmax and ΔSUVmean after one cycle showed consistent associations with pathologic outcome within this cohort. Correlation of SUV changes with biological markers of response Baseline Ki-67 scores and TIL levels (Fig. 4A and B) were available for 13 patients in the FDG group. Data were missing for one patient due to insufficient viable tumor content in the biopsy sample. Ki-67 was significantly positively correlated with baseline SUVmax at all time points (r = 0.648, 0.670, and 0.713 at 90, 120, and 180 minutes, respectively; all P < 0.02), with a similar trend observed for SUVmean. In contrast, baseline TIL levels were not significantly associated with SUVmax/mean at any imaging time point (all P > 0.75; Fig. 4C; Supplementary Table S6). Correlation of SUV parameter changes with biological markers of response. A, Representative H&E–stained tumor sections demonstrating a spectrum of stromal TIL levels: low (1%–10%), moderate (10%–40%), and high (>40%) infiltration. B, Representative IHC images showing high and low Ki-67 expression levels. C, Spearman correlation matrix showing associations between baseline tissue biomarkers (Ki-67 and TILs) and FDG-PET parameters (SUVmax and SUVmean at 90, 120, and 180 minutes). Symbol color and size indicate the strength and direction of each correlation. P values are shown. D, Paired imaging–pathology case example from a patient who achieved a pCR. Left, baseline. Right, after cycle 1. Top, FDG-PET/CT images (arrow = index lesion) showing >70% reduction in SUVmax and SUVmean following one cycle of NACT. Middle, Ki-67 IHC demonstrating a decrease in the proliferative index from 85.1% at baseline to 65.7% after cycle 1. Bottom, H&E-stained sections showing an increase in stromal TILs from 10% at baseline to 50% after one cycle, consistent with enhanced immune infiltration. E, Spearman correlation matrix showing associations between percentage change in tissue biomarkers (ΔKi-67 and ΔTILs) after one cycle of therapy, early metabolic PET response (ΔSUVmax and ΔSUVmean), and final pathologic outcome (RCB index). P values are shown. F, Baseline tissue biomarkers according to RCB score. The Fisher exact test was used to compare proportional values, and the Mann–Whitney test for comparing median values within the two response groups. C1, cycle 1. To evaluate whether treatment-induced changes in tissue biomarkers could predict pathologic response, we analyzed post–cycle 1 changes in Ki-67 (ΔKi-67) and TILs (ΔTILs) and compared these with ΔSUVmax/mean and RCB score. Paired data were limited to a subset of patients (11 patients for Ki-67 and 10 patients for TILs) due to constraints in tissue quality and insufficient viable tumor content in some samples (Fig. 4D and E; Supplementary Table S7). No significant correlations were observed between ΔKi-67 and ΔSUVmax/mean nor between ΔKi-67 and RCB index, suggesting that changes in proliferation, at least as measured by ΔKi-67, did not reflect treatment effect sufficiently to correlate with RCB category in this cohort (Fig. 4E). However, ΔTILs demonstrated notable inverse correlations with both metabolic response and final RCB index. Specifically, ΔTILs correlated significantly with ΔSUVmax at 120 minutes (r = −0.71; P = 0.02), with ΔSUVmean at 120 (r = −0.685; P = 0.03) and 180 minutes (r = −0.683; P = 0.04), and with final RCB index (r = −0.68; P = 0.03; Fig. 4E; Supplementary Table S7). These findings suggest that early increases in immune activation, reflected by increased TILs, may be associated with reduced tumor metabolic activity and lower residual disease at surgery. In contrast to these treatment-induced changes, baseline Ki-67 and TIL levels were not significantly associated with pathologic response (Fig. 4F; Supplementary Table S6). Though limited by small sample size, this pilot analysis highlights the potential utility of integrating imaging and immune biomarker dynamics for early response prediction in TNBC. Safety Both PET tracers were well tolerated, with no serious or unexpected adverse events. Discussion This prospective phase II study demonstrates that early metabolic response assessment with FDG-PET after a single cycle of NACT has clinical validity in TNBC. The prespecified primary endpoint—correlation between early ΔSUV and post–cycle 3 MRI response—was fulfilled. Within this cohort, metabolic response also showed stronger associations with pathologic outcome than RECIST-based MRI, although the study was not designed or powered for formal modality comparison. Our findings are consistent with prior studies reporting the prognostic relevance of PET response after 1 to 2 NACT cycles (15, 16, 19, 33, 34) but extend the literature in two important ways. First, this trial prospectively evaluated metabolic response using RCB, the internationally standardized and clinically validated pathologic endpoint now used to select patients for escalation in post-NACT therapy trials. Although our pragmatic dichotomization of RCB into favorable (RCB-0/I) and unfavorable (RCB-II/III) groups does not capture the full prognostic granularity of the original system, we also analyzed the relationship between early metabolic response and the continuous RCB index, a clinically validated pathologic endpoint. The observed correlation between post–cycle 1 metabolic change and the continuous RCB index demonstrates that this association exists across the full spectrum of residual disease and is not an artefact of the binary grouping. Second, this study suggests that biological response on PET may precede detectable morphologic change on anatomic imaging. A ≥45% reduction in SUVmax/SUVmean after one cycle identified patients achieving RCB-0/I. This cohort-specific threshold is substantially higher than the EORTC-recommended 15% to 25% metabolic response cutoff (13). In TNBC, in which baseline FDG uptake is typically higher and metabolic variability greater than in other breast cancer subtypes, a larger relative reduction may be required to distinguish responders from nonresponders (16, 19, 35). Consistent with this, studies evaluating early FDG-PET response in mixed breast cancer populations, including hormone receptor–positive and HER2-positive disease, have generally reported lower metabolic response thresholds. The higher threshold observed in our TNBC cohort is therefore biologically plausible and aligns with the greater baseline FDG avidity and dynamic metabolic response characteristic of this subtype. Nevertheless, optimal metabolic response thresholds remain uncertain and require prospective validation in larger, subtype-specific cohorts. Accurate prediction of response after a single NACT cycle carries significant clinical implications. Early identification of nonresponders could enable timely treatment escalation, whereas strong metabolic responders may be candidates for de-escalation. In our study, midtreatment MRI response was not significantly associated with RCB, which may reflect sample-size constraints and highlights ongoing uncertainty regarding the ability of anatomic response assessment alone to predict pathologic outcome after NACT (36). Our findings support the formal testing of FDG-PET as an early-response monitoring tool in TNBC (37). A key strength of this study is the analytic validation performed in part A. To our knowledge, this represents the first TNBC-specific PET repeatability analysis, establishing SUVmax and SUVmean as reliable and reproducible parameters for response monitoring. We also evaluated multiple postinjection imaging time points, demonstrating no added value beyond the standard 90-minute acquisition, supporting feasibility and clinical implementation. A novel biological insight from this study is the association between early metabolic response and dynamic immune activation. Baseline stromal TILs—a recognized prognostic and predictive biomarker in TNBC—did not correlate with PET uptake or response, likely reflecting sampling variability and tumor heterogeneity. However, increases in TILs after one cycle strongly correlated with metabolic response and RCB. Although this finding must be interpreted with caution due to the limited number of paired samples and inherent challenges of biopsy-based sampling variability, this correlation is biologically plausible and suggests that chemotherapy-induced immune mobilization contributes to the early metabolic collapse measured by FDG-PET. Thus, ΔSUV may serve not only as a marker of tumor cell kill but also as a surrogate for favorable microenvironmental immune reprogramming. This metabolic–immune coupling offers a unique window for future adaptive designs integrating PET imaging with dynamic immune biomarkers. Additional methodologic insights emerged. SUVpeak and SULpeak demonstrated inferior per-patient repeatability, likely limiting their practicality in early-stage TNBC. Although PERCIST recommends lean body mass correction, fluctuation in body habitus was minimal in this cohort, and SULpeak offered no advantage. Together, these findings support SUVmax and SUVmean as the most robust PET biomarkers for early response assessment in TNBC. The evolving treatment landscape of TNBC, particularly the incorporation of immune checkpoint inhibitors (ICI) into neoadjuvant regimens (38), raises important considerations. Immune-mediated metabolic changes—such as increased FDG uptake from activated immune infiltrates—may influence SUV measurements and complicate response interpretation. Recent work by Seban and colleagues (33) suggests that pretreatment SUVmax and total metabolic tumor volume predict pathologic response in TNBC, although optimal cutoffs differed between NACT-alone and NACT plus pembrolizumab cohorts. Whether conventional SUV thresholds remain valid in the immunotherapy era requires further evaluation. In addition, platinum agents were not routinely incorporated into the NACT backbone during the conduct of this study. Carboplatin is now frequently included in contemporary TNBC regimens and is associated with higher pCR rates. Differences in tumor cell kill kinetics and treatment-induced metabolic changes may influence the magnitude and timing of FDG uptake reduction; therefore, the metabolic response thresholds identified in this cohort should be validated in patients receiving platinum-containing and immunotherapy-based regimens. Although pseudoprogression has not been systematically studied in TNBC, data from melanoma and lung cancer suggest that ICI-related immune infiltration—and its impact on FDG uptake—typically occurs later in treatment, often beyond two–three cycles. Use of FDG-PET immediately after one cycle may therefore offer a key advantage, capturing early tumor response before such immune-related effects become prominent. This timing could mitigate the risk of misinterpreting immune activation as tumor progression and supports pre–cycle 2 as a robust window for response assessment in future immunotherapy-containing regimens. Similarly, the potential for PET-MRI fusion to offer a multiparametric approach to assess tumor biology—including metabolic activity, cellularity (via DWI), and spatial anatomic detail (39)—represents an important avenue for future research. This approach may be enhanced through the use of immune-PET tracers as complementary biomarkers to offer valuable insights into dynamic immune responses. Additionally, analysis of texture features derived from PET images can further preserve spatial complexity and may reflect underlying tumor–immune interactions more sensitively than standard imaging metrics (40). Collectively, these emerging imaging strategies have the potential to inform the development of multiparametric models for more accurate response prediction in TNBC, particularly in the context of ICI-based regimens. Several limitations merit consideration. The sample size was below the prespecified target for the primary endpoint, limiting statistical power and generalizability. Accordingly, the findings should be considered exploratory and hypothesis-generating and require prospective validation in larger, multicentre cohorts. The FLT group, although exploratory, was underpowered to assess its full utility. FLT, with its specificity for DNA synthesis, may yet prove valuable in immunotherapy-treated patients by improving discrimination between tumor response and inflammatory changes. Additionally, the change in chemotherapy sequencing (from T-EC to EC-T) during the study meant that different agents were administered prior to the post–cycle 1 PET scan. As taxanes and anthracyclines differ in tumor cell kill kinetics and metabolic effects, early ΔSUV could theoretically be influenced by the specific agent received. Furthermore, assessment of a single index lesion may not fully capture intrapatient tumor heterogeneity, particularly in node-positive disease. Although the correlation between early metabolic change and final RCB score (a sequence-agnostic endpoint) supports the robustness of the observed association, these sources of heterogeneity warrant validation in more uniform cohorts. Although a single-reader approach ensured consistency in VOI delineation across scans, reproducibility of these measurements should be confirmed in future multicenter studies. In addition, eligibility in this study was restricted to women. Although TNBC can occur in men, male breast cancer is rare, and no male patients were encountered during the recruitment period. Consequently, the generalizability of these findings to male TNBC requires further investigation. Finally, early PET-based response assessment may offer clinical and health economic benefits by reducing exposure to ineffective therapy and associated toxicities. Future trials should explore the clinical utility, cost-effectiveness, and feasibility of incorporating early FDG-PET into response-adapted neoadjuvant treatment pathways for TNBC. In conclusion, FDG-PET after a single NACT cycle was associated with both subsequent radiologic and pathologic response in TNBC. The metabolic response threshold identified here is exploratory and requires prospective validation. If validated in larger trials, including those assessing immunotherapy-containing regimens, treatment escalation or de-escalation based on early PET-based assessment could be explored as a tool to ultimately improve patient outcomes. Supplementary Material Supplementary Figure 1: Repeatability of FDG tracer at exploratory timepoints and Predictive Performance of the FLT tracer. A–B. Bland–Altman plots of log-transformed SUV measurements from the two baseline repeatability scans for the FDG tracer at 120 (A) and 180 minutes (B) post-injection. Each point represents an individual lesion. Log SUVmax, SUVmean, SUVpeak and SULpeak are shown. The gray shaded band (log ±0.14) represents the ±15 % variability threshold used to assess test–retest repeatability. C. Individual FDG-PET responses after one cycle of neoadjuvant chemotherapy (NACT) for SUVpeak and SULpeak parameters at 90, 120, and 180 minutes post-injection. Bars are colored by post-cycle 3 MRI response category (complete response, partial response, stable disease, or progressive disease) and annotated according to patient (Pt) number. D. Individual FLT-PET responses for SUVpeak and SULpeak at 90 minutes post-injection. *Patient 7 underwent an early MRI after two cycles due to clinical progression. Supplementary Tables 1-8.