Machine learning for the prediction of three-year survival in locally advanced breast cancer patients receiving neoadjuvant chemotherapy using quantitative ultrasound imaging.
Authors
Affiliations (8)
Affiliations (8)
- Department of Physics, Toronto Metropolitan University, Toronto, ON, Canada. [email protected].
- Institute for Biomedical Engineering, Science and Technology (iBEST), Keenan Research Centre for Biomedical Science, St. Michael's Hospital, Toronto, ON, Canada. [email protected].
- Department of Radiation Oncology, Sunnybrook Health Sciences Centre, Toronto, ON, Canada.
- Department of Radiation Oncology, University of Toronto, Toronto, ON, Canada.
- Physical Sciences, Sunnybrook Research Institute, Toronto, ON, Canada.
- Department of Medical Biophysics, University of Toronto, Toronto, ON, Canada.
- Department of Physics, Toronto Metropolitan University, Toronto, ON, Canada.
- Institute for Biomedical Engineering, Science and Technology (iBEST), Keenan Research Centre for Biomedical Science, St. Michael's Hospital, Toronto, ON, Canada.
Abstract
Locally Advanced Breast Cancer (LABC) is a serious type of cancer with a poor prognosis despite advances in cancer treatment. As the disease is often inoperable, current guidelines recommend upfront aggressive neoadjuvant chemotherapy (NAC). While conventional ultrasound provides tissue echogenicity data, comparing images remains challenging due to the varied hardware configurations and instrument settings. Quantitative ultrasound (QUS) corrects this by using normalized power spectra calculations to derive quantitative parameters that are independent of instrument settings. In this work, we present an integrated deep-learning pipeline that reduces the possibility of data leakage and facilitates efficient data processing by combining scaling, oversampling, feature selection, and classification into a single framework. The pipeline is used to predict the three-year survival in LABC patients receiving NAC using QUS imaging before treatment initiation. The pipeline was trained on five quantitative ultrasound maps at the pre-treatment stage. The average acoustic concentration was the most predictive feature, achieving a recall and precision of 95% and 91%, respectively, for the survivor class. This work demonstrates that QUS may be used as a non-invasive biomarker for differentiating between LABC survivors and non-survivors at the pre-treatment stage. Prediction of the three-year survival rates of LABC patients before treatment can be used for prognosis, treatment planning, and patient decision-making.