Multiscale Fusion Models With Genomic, Topological, and Pathomic Features to Predict Response to Radiation Therapy...

Area of Science:

• Computational biology
• Medical imaging analysis
• Oncology

Background:

• Personalized radiation therapy requires accurate prediction of treatment response.
• Histopathology images offer a rich source of data for predicting radiosensitivity.
• Integrating multiscale features may enhance predictive accuracy.

Purpose of the Study:

• To investigate the efficacy of artificial intelligence (AI) fusion models using multiscale features for predicting radiation therapy response in non-small cell lung cancer (NSCLC).
• To evaluate the predictive power of genomic, pathomic, and topological features, individually and in combination.

Main Methods:

• Development of base models using genomic, pathomic, and topological features from histopathology images of NSCLC patients.
• Construction of fusion models combining these multiscale features.
• Validation of models using internal and external test datasets from TCGA and CPTAC cohorts.

Main Results:

• Topological models demonstrated superior classification and prognostic prediction compared to genomic and pathomic models alone.
• The best fusion model, incorporating genomic, topological, and pathomic features, achieved the highest AUC for predicting treatment response (0.846 internal, 0.731 external).
• Inner-cell topological structure appears to contain radiosensitivity-related information.

Conclusions:

• Multiscale AI fusion models, particularly those incorporating topological features, can effectively predict radiation therapy response in NSCLC.
• These models hold potential for assisting clinicians in selecting patients for personalized radiation therapy, improving outcomes.
• The findings suggest that topological analysis of histopathology images is a valuable tool in precision oncology.