The value of combined assessment of tumor cellularity and metabolism by PET/MRI in predicting pathological complete response after neoadjuvant treatment in breast cancer

Abstract Purpose This study aimed to evaluate the role of primary tumor quantitative parameters obtained from 18 F-FDG PET/MRI in breast cancer (BC) patients in the prediction of pathological complete response (pCR). Methods Patients with BC who underwent PET/MRI for staging and neoadjuvant treatment (NAT) response evaluation, and underwent surgical resection at the end of the treatment (EOT) were retrospectively reviewed. A total of 59 patients were included in the study. Maximum standard uptake value (SUVmax), metabolic tumor volume (MTV), total lesion glycolysis (TLG) and minimum apparent diffusion coefficient (ADCmin) value of the primary tumors were obtained from the initial and EOT PET/MRI. The patients were divided into two groups according to PERCIST: those with complete metabolic response (CMR) and those without CMR. The role of quantitative parameters in predicting pCR was evaluated by using decision tree model as a result of classifications in the Rweka package of the R software. Results pCR was detected in 22 of 59 patients (37.3%). The sensitivity, specificity and accuracy of the evaluation according to PERCIST in predicting pCR were found to be 90.9%, 75.7% and 81.4%, respectively. In the decision tree model, patients were classified using a two-layer model with the values of EOT_ADCmin (cutoff value > 1.6 × 10 − 3 mm 2 /s, complete response) and EOT_SUVmax (cutoff value > 1.45, incomplete response). According to the confusion matrices, the model correctly classified 56 patients (sensitivity: 95.5%, specificity: 94.6%, accuracy: 94.9%). The accuracy of the model was found to be 93.2% with the usage of 10-fold cross-validation method. Conclusion In BC patients who received NAT, EOT_ADCmin and EOT_SUVmax were found as predictive factors for pCR. The combined assessment of tumor metabolism and cell density may be useful in non-invasive prediction of pCR, with higher accuracy compared to PERCIST.