Transfer Learning Approach to Vascular Permeability Changes in Brain Metastasis Post-Whole-Brain Radiotherapy.

Area of Science:

• Biomedical Imaging
• Machine Learning in Oncology
• Translational Research

Background:

• Brain metastasis (BM) models are crucial for studying tumor vascular permeability.
• Dynamic Contrast-Enhanced Magnetic Resonance Imaging (DCE MRI) is vital for assessing vascular properties.
• Previous Convolutional Neural Network (CNN) models showed promise in glioma studies.

Purpose of the Study:

• To validate a previously developed CNN for brain metastasis (BM) using transfer learning in a new small animal model.
• To assess the CNN's ability to predict vascular permeability in multifocal intracranial metastases.
• To evaluate the CNN's performance in detecting radiotherapy-induced changes in vascular permeability.

Main Methods:

• Transfer learning was employed, adapting a pre-trained CNN from a glioblastoma model to BM DCE MRI datasets.
• The CNN was re-trained to correlate BM DCE images with permeability maps from the Extended Tofts Model (ETM).
• The model's predictive accuracy and spatial correlation with ETM pharmacokinetic (PK) maps were evaluated. Further testing involved assessing radiotherapy (WBRT) effects on vascular permeability.

Main Results:

• The transferred CNN accurately predicted BM vascular permeability, showing excellent spatial correlation with ETM PK maps.
• The CNN detected significantly increased Ktrans in WBRT-treated tumors (p < 0.01), consistent with ETM findings.
• The model demonstrated robust performance in characterizing permeability changes induced by radiotherapy.

Conclusions:

• The developed CNN serves as an efficient and accurate tool for characterizing vascular permeability in small animal brain tumor models.
• This approach facilitates the assessment of treatment responses in preclinical brain tumor research.
• Transfer learning enhances the utility of CNNs for analyzing complex imaging data in oncology.