Artificial intelligence and radiomics biomarkers for treatment response prediction in advanced HER2-negative breast cancer

Abstract

Study Aim

Despite the development of novel therapies, breast cancer mortality remains high. Improved treatment response assessment tools are needed. We aimed to test the transferability of externally validated AI/radiomics biomarkers for predicting treatment response in advanced, hormone receptor-positive (HR+), HER2-breast cancer treated with xentuzumab, exemestane, and everolimus.

Methods

Patient data from a phase Ib/II trial (May 2014–October 2016) were analyzed retrospectively. Eight imaging biomarkers (liver and overall tumor volume, two radiomics features representing tumor heterogeneity at both baseline and week 8) were validated for predicting clinical benefit using AUC analysis. An ancillary AI analysis developed a signature for predicting best overall response using 40 variables (3 clinical and 37 imaging) in the same cohort.

Results

Of 106 patients with data available for analysis, 28 had no clinical benefit from treatment (Group A) vs. 78 with clinical benefit (Group B). Seven of eight imaging biomarkers demonstrated significant predictive value. Participants in Group B exhibited significantly lower baseline and follow-up measures of liver and overall tumor volume, alongside marked changes in tumor heterogeneity by week 8. In our ancillary AI/radiomics model, the dominant drivers of prediction were changes in both liver tumor and overall tumor volume during treatment and the number of osteoblastic lesions on baseline bone scans.

Conclusion

This cross-cancer proof-of-concept testing study supports the feasibility of applying multimodal AI/radiomics biomarkers to predict treatment response in advanced HR+, HER2− breast cancer, laying the foundation for broader pancancer and pantreatment applications pending further validation.

Clinical trial registration number

NCT02123823