Ultrasound localisation microscopy predicts neoadjuvant chemotherapy response in patients with breast cancer.

BACKGROUND: Patients with breast cancer undergoing neoadjuvant chemotherapy mostly show insufficient therapy response and would highly benefit from an early prediction of pathological complete response (pCR). As vessels play a major role in drug delivery, ultrasound localisation microscopy (ULM) holds great promise as predictive tool by visualising vascular networks in high detail, at a resolution beyond the acoustic diffraction limit. METHODS: In this prospective pilot study (NCT05445050), we examined patients with breast cancer undergoing neoadjuvant chemotherapy using a ULM algorithm designed for conventional clinical ultrasound systems. To generate a vascular fingerprint, we performed the most detailed radiomics ULM analysis to date, based on longitudinal data from 20 patients. Features were used to classify therapy response using logistic regression and the performance of ULM was compared to clinically available imaging methods. FINDINGS: Compared to patients with non-pCR (n = 12), patients with pCR (n = 8) showed a more pronounced vascular network before therapy. Furthermore, their vessel coverage strongly changed following the first chemotherapy administration (26.97 [20.23-36.24] % vs. 6.00 [2.73-10.49] %), representing a promising biomarker for therapy prediction (p < 0.01, effect size: 2.21). ULM features allowed to correctly predict up to 85% (AUC: 0.85) of patient responses and already 75% (AUC: 0.73) before the start of therapy, thus outperforming tumour size measurements (accuracy up to 60%, AUC up to 0.53), B-mode images (accuracy up to 75%, AUC up to 0.76), maximum intensity projections of contrast-enhanced ultrasound (accuracy up to 75%; AUC up to 0.79), and histological CD31 staining (accuracy: 75%; AUC: 0.83). INTERPRETATION: Our findings emphasise the importance of vascular assessment in predicting pCR in patients with breast cancer and promote the translation of ULM as powerful clinical tool with the potential to optimise neoadjuvant treatment schedules. FUNDING: This work was supported by the German Research Foundation.