FDG-PET/CT and Multimodal Machine Learning Model Prediction of Pathological Complete Response to Neoadjuvant...

Area of Science:

• Oncology
• Medical Imaging
• Machine Learning

Background:

• Triple-negative breast cancer (TNBC) is aggressive and heterogeneous, with outcomes linked to neoadjuvant chemotherapy (NAC) response.
• Pathological complete response (pCR) after NAC correlates with improved patient survival in TNBC.
• Predicting pCR before treatment is crucial for tailoring therapy and improving outcomes.

Purpose of the Study:

• To develop and validate a machine learning (ML) algorithm for predicting pCR in TNBC patients.
• To identify key predictors of pCR from multimodal baseline data.
• To explore the association between predicted pCR and event-free survival (EFS).

Main Methods:

• A cohort of 57 TNBC patients undergoing FDG-PET/CT before NAC was analyzed.
• An ML algorithm was trained using 241 predictors: clinical, histopathological, genomic, and PET radiomic features.
• Predictive performance was evaluated using the area under the ROC curve (AUC); EFS was analyzed with Kaplan-Meier.

Main Results:

• The best ML model achieved an AUC of 0.82 for pCR prediction.
• Predictors included a combination of PET, radiomic, histopathological, genomic, and clinical features, emphasizing multimodal data importance.
• Patients predicted to have pCR showed a trend towards longer EFS, though not statistically significant (p=0.09).

Conclusions:

• ML models using baseline multimodal data can effectively predict pCR in TNBC patients undergoing NAC.
• This predictive capability may help in identifying patients who could benefit from intensified treatment or immunotherapy.
• The study highlights the potential of integrating diverse data types for personalized TNBC management.