Machine Learning Models Based on Stretched-Exponential Diffusion Weighted Imaging to Predict TROP2 Expression in Nude...

Area of Science:

• Oncology
• Radiology
• Machine Learning

Background:

• Trophoblast cell surface antigen 2 (TROP2) is a key target in breast cancer therapy.
• Noninvasive methods to assess TROP2 expression are crucial for treatment planning.
• This study investigates predicting TROP2 expression using advanced imaging techniques.

Purpose of the Study:

• To evaluate the efficacy of machine learning models based on multi-b-value diffusion-weighted imaging (DWI) with the stretched-exponential model (SEM) for predicting TROP2 expression in breast cancer.
• To compare the performance of extreme gradient boosting (XGBoost), logistic regression, and adaptive boosting (AdaBoost) classifiers.
• To identify key imaging parameters for accurate TROP2 expression prediction.

Main Methods:

• Thirty-two nude mouse breast cancer models underwent 1.5T MRI using multi-b-value DWI.
• Stretched-exponential model (SEM) parameters (DDC, α) and histogram features were extracted using FireVoxe software.
• TROP2 expression was quantified via immunohistochemical staining (IOD); machine learning models were trained and validated.

Main Results:

• Eight key imaging parameters were selected from 38 extracted features.
• The XGBoost model achieved the highest Area Under the Curve (AUC) of 0.828 on the validation set.
• XGBoost demonstrated excellent performance in training (AUC=1) and test sets (AUC=1), with good calibration and favorable net benefits on decision curve analysis.

Conclusions:

• Machine learning models utilizing SEM-based DWI show significant potential for predicting TROP2 expression in breast cancer.
• The XGBoost classifier exhibited superior predictive accuracy, highlighting its promise for clinical applications.
• This noninvasive approach could aid in personalized breast cancer treatment strategies.