255P - Predicting early recurrence in breast cancer patients undergoing neo-adjuvant chemotherapy through MRI-radiomics analysis

Background

Breast cancer (BC) is a major global health concern. Neoadjuvant treatment (NAT) is standard for many cases, aiming to shrink tumors before surgery. Despite its effectiveness, some patients still experience early recurrence (ER), defined as relapse within the 3 years from the start of NAT, impacting their survival. Efforts are being made to predict who is at higher risk of ER using tools like breast MRI and radiomics, which extract information from images. Predicting ER after NAT remains a challenge. The study investigates the potential of radiomics analysis, combined with clinical and radiological variables, in predicting ER.

Methods

Patients with BC who underwent staging MRI, NAT, and surgery at our Center (from 2012 to 2021) were included. Clinical variables evaluated included pathological complete response (pCR), ER, and tumor subtype. Radiological variables included tumor response according to RECIST criteria. Four breast radiologists reviewed breast MRI, annotated regions of interest and extracted radiomics features. Pure radiomics models and combined models (clinical-radiological, radiological-radiomics, and clinical-radiomics) were developed. The area under the curve (AUC) was calculated for each model, and the models were compared in terms of accuracy, sensitivity, and specificity.

Results

A total of 211 patients were included, the ER prevalence was of 11.34%. Patients with pCR or partial response to NAT and Luminal tumor subtype had a lower likelihood of developing ER (p = 0.001 and p = 0.037, respectively). Two radiomics features were statistically significant associated with ER: F_cm_2.5D.energy and F_cm_2.5D.joint.entr. The AUC values for combined models were 0.77 (radiological-radiomics model), 0.68 (clinical-radiomics model), and 0.74 (clinical-radiological model). The radiological-radiomics model was significantly more accurate in predicting ER than the pure radiological and radiomics models (p<0.001 and p<0.03, respectively).

Conclusions

The radiological-radiomics model, combining radiomics features and RECIST criteria, showed the most promising results in predicting ER.

Clinical trial identification

ID 6081 26.10.2023.

Editorial acknowledgement

Legal entity responsible for the study

Anna D'Angelo.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.