JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

18F-FDG PET/CT-based deep radiomic models for enhancing chemotherapy response prediction in breast cancer

  • 0Advanced Molecular Imaging in Radiotherapy (AdMIRe) Research Lab, School of Health Sciences, College of Health and Human Sciences, Purdue University, West Lafayette, IN, 47907, USA.
Medical Oncology +

|

August 11, 2025

|

August 11, 2025

View abstract on PubMed

Summary

This summary is machine-generated.

Deep radiomic models accurately predict breast cancer chemotherapy response after the first cycle. Integrating deep learning features with XGBoost improves early treatment assessment and personalized therapy strategies.

Area Of Science

  • Oncology
  • Medical Imaging
  • Artificial Intelligence

Background

  • Accurate prediction of breast cancer chemotherapy response is crucial for personalized treatment.
  • Early identification of non-responders can guide therapeutic strategy adjustments.
  • Radiomics and deep learning offer potential for enhanced predictive accuracy.

Purpose Of The Study

  • To develop and evaluate deep radiomic models for predicting chemotherapy response in breast cancer patients after the first treatment cycle.
  • To compare the performance of XGBoost, random forest, logistic regression, and support vector machine models.
  • To assess the impact of integrating deep learning-derived features on predictive accuracy.

Main Methods

  • Retrospective analysis of 18F-Fludeoxyglucose PET/CT imaging and clinical data from 60 breast cancer patients.
  • Extraction of radiomic features from the gross tumor volume (GTV) delineated on PET images.
  • Application of a Squeeze-and-Excitation Network (SENet) deep learning model to generate additional features.
  • Development and comparison of XGBoost, random forest, logistic regression, and support vector machine models using radiomic and deep learning features.
  • Performance evaluation using receiver operating characteristic area under the curve (ROC AUC) and fivefold cross-validation.

Main Results

  • The XGBoost model achieved the highest AUC (0.85) using only radiomic features.
  • Incorporating deep learning-derived features from SENet significantly improved AUC values for all models (XGBoost: 0.92, RF: 0.88, LR: 0.90, SVM: 0.61).
  • The enhanced models enabled early prediction of chemotherapy response based on pre-treatment (T1) and post-first-cycle (T2) imaging data.

Conclusions

  • Integrating deep learning-derived features substantially enhances the predictive performance of radiomic models for breast cancer chemotherapy response.
  • The XGBoost model, augmented with deep learning features, demonstrates superior capability for early and accurate prediction of treatment outcomes.
  • These findings support the potential of advanced radiomic models in optimizing personalized therapeutic strategies for breast cancer patients.

Keywords: