PET/CT-based quantification of intratumoral heterogeneity for predicting pathological complete response in breast cancer patients treated with neoadjuvant chemotherapy: A multicentre study.

Abstract

PURPOSE: To develop a quantitative measure of intratumoral heterogeneity (ITH) from pretreatment 18F-fluorodeoxyglucose positron emission tomography/computed tomography (18F-FDG PET/CT) images and to evaluate its potential for predicting pathological complete response (pCR) to neoadjuvant chemotherapy (NAC) in patients with breast cancer (BC). MATERIALS AND METHODS: This retrospective study included patients with BC who underwent NAC at four hospitals between 2013 and 2025. Tumour regions on pretreatment 18F-FDG PET/CT images were clustered to extract intratumoral ecological diversity features, while radiomic features were extracted from the entire tumour. Output probabilities from imaging-based extreme gradient boosting (XGBoost) models were used to generate an ITH index and a radiomics score, respectively. Logistic regression analysis was performed to identify variables associated with pCR. Significant variables, including clinicopathologic characteristics, radiomics score, and ITH index, were integrated into a combined predictive model (ICR-COME). Model performance and clinical utility were evaluated using receiver operating characteristic, calibration curve, and decision curve analyses (DCA). Survival differences between predicted pCR and non-pCR groups were assessed using Kaplan–Meier analysis. RNA sequencing data from 12 patients were analysed to further explore the biological basis of ICR-COME. RESULTS: The study included 323 women (median age, 51 years ± 10[SD]). For predicting pCR in BC patients, ICR-COME achieved the area under the receiver operating characteristic curve (AUC) values of 0.95 in the training cohort and 0.87 and 0.88 in external test cohorts B and C, respectively, which were significantly superior to those of the clinicopath model and the conventional radiomics model using 18F-FDG PET/CT (all P < 0.05, DeLong test). Calibration curve and DCA indicated that ICR-COME achieved satisfactory concordance and provided the greatest clinical net benefit. Patients in all three cohorts predicted as non-pCR exhibited poorer prognosis (P < 0.05, log-rank test). Furthermore, RNA sequencing and multiplex immunofluorescence analyses revealed that high ICR-COME scores were associated with increased immune cell infiltration and enhanced activation of immune-related pathways. CONCLUSIONS: ICR-COME, which integrates the ITH index, the radiomics score, and clinicopathologic variables, demonstrated strong predictive performance for pCR to NAC and prognosis in BC patients.