An Ultrasound-based Machine Learning Model for Predicting Tumor-Infiltrating Lymphocytes in Breast Cancer

Area of Science:

• Oncology
• Medical Imaging
• Machine Learning

Background:

• Tumor-infiltrating lymphocytes (TILs) are crucial prognostic indicators in breast cancer (BC), influencing immune response.
• Accurate prediction of TIL levels is vital for tailoring personalized treatment strategies in BC management.

Purpose of the Study:

• To develop and validate machine learning models for predicting TIL levels in breast cancer.
• To utilize ultrasound-derived radiomics and clinical features for enhanced TIL prediction accuracy.

Main Methods:

• Retrospective analysis of 256 breast cancer patients with ultrasound imaging.
• Extraction of 1712 radiomics features from intratumor and peritumor regions; Boruta method for feature selection.
• Development of radiomics, clinical, and combined radiomics-clinical (R-C) models using Extra Trees Classifier.

Main Results:

• The combined R-C and radiomics models significantly outperformed the clinical model in predicting TIL levels (AUC 0.869/0.838 vs 0.627).
• Five key radiomics features (4 peritumor, 1 intratumor) were identified as significant predictors.
• Superior accuracy and calibration were observed for the R-C and radiomics models, with the R-C model showing the highest net benefit in decision curve analysis.

Conclusions:

• Ultrasound-derived radiomics is an effective tool for predicting TIL levels in breast cancer.
• This predictive capability offers valuable insights for developing personalized treatment and surveillance strategies in BC.