Multiomics-Based Outcome Prediction in Personalized Ultra-Fractionated Stereotactic Adaptive Radiotherapy (PULSAR).

Area of Science:

• Radiology
• Oncology
• Medical Imaging Analysis

Background:

• Brain metastasis (BM) treatment response prediction is crucial for patient management.
• Pulsed Radiofrequency (PULSAR) is a treatment modality for BM.
• Integrating diverse data types can enhance predictive accuracy.

Purpose of the Study:

• To develop and evaluate a multiomics approach for predicting treatment response in BM patients undergoing PULSAR.
• To integrate radiomics, dosiomics, and delta features for enhanced prediction.
• To assess the performance of an ensemble feature selection (EFS) model.

Main Methods:

• Retrospective analysis of 39 BM patients (69 lesions) treated with PULSAR.
• Extraction of radiomics, dosiomics, and delta features from pre- and intra-treatment MRI and dose distributions.
• Evaluation of six individual models and an EFS model for predicting >20% volume reduction.

Main Results:

• The EFS model, integrating multiomics features, outperformed individual models.
• EFS model achieved an AUC of 0.979, accuracy of 0.917, and F1 score of 0.821.
• Top predictive features included post-wavelet transformed and original image features.

Conclusions:

• A data-driven multiomics approach is feasible for predicting PULSAR treatment outcomes in BM.
• Integrating multiomics with intra-treatment decision support can optimize patient management.
• This approach may reduce risks of under- or over-treatment in BM patients.