A Radiogenomic Deep Ensemble Learning Model for Identifying Radionecrosis Following Brain Metastases (BM)...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Stereotactic radiosurgery (SRS) is a common treatment for brain metastases (BM).
• Distinguishing post-SRS radionecrosis from tumor recurrence noninvasively remains a clinical challenge.
• Accurate differentiation is crucial for effective patient management and treatment planning.

Purpose of the Study:

• To develop a deep ensemble learning model for identifying radionecrosis versus recurrence in patients with BM after SRS.
• To integrate patient clinical features and genomic profiles with imaging data for improved diagnostic accuracy.
• To address the limitations of current noninvasive imaging methods in differentiating these conditions.

Main Methods:

• A deep neural network (DNN) was trained on 3-month post-SRS T1+c MRI scans from 90 BMs (62 patients) with non-small cell lung cancer.
• An ensemble model fused DNN-extracted deep features with clinical (D+C) or genomic (D+G) data using a novel positional encoding method.
• Model performance was evaluated against an image-only DNN and the individual D+C and D+G submodels.

Main Results:

• The deep ensemble model achieved a high diagnostic performance (ROC AUC = 0.91 ± 0.04).
• This significantly outperformed the image-only DNN (ROC AUC = 0.71 ± 0.05) and individual submodels (D+C: 0.82 ± 0.03, D+G: 0.83 ± 0.02).
• The model demonstrated strong sensitivity (0.87 ± 0.16), specificity (0.86 ± 0.08), and accuracy (0.87 ± 0.04).

Conclusions:

• The deep ensemble model shows superior performance in differentiating BM radionecrosis from recurrence.
• Integrating multi-modal data (imaging, clinical, genomic) enhances diagnostic capabilities.
• This AI-driven approach holds significant potential for improving clinical decision-making in BM management.