IMG-112. Color-coded change maps for tracking metastatic brain cancer progression on longitudinal magnetic resonance imaging

Abstract BACKGROUND Tracking metastatic brain cancer progression over time presents significant clinical challenges. Accurate detection of brain metastases on magnetic resonance imaging (MRI) is essential for determining correct tumor resection margins and distinguishing tumor progression from treatment response. METHODS We adapted the VALE method—a satellite imaging technique that generates color-coded change maps—to highlight tumor margins in longitudinal brain MRI scans. We used the publicly available NYUMets dataset (1,429 patients with stage IV metastatic cancer, each with on average 6 longitudinal MRIs including T1-weighted pre-contrast, T1-weighted post-contrast, and FLAIR sequences), and selected three timepoints approximately six months apart. Our processing pipeline involved four key steps: (1) registering FLAIR and T1 post-contrast images to the baseline T1 pre-contrast scan, (2) normalizing median pixel intensities across timepoints using white matter intensity from the baseline scan, (3) performing histogram clipping, and (4) creating color-coded change maps by overlaying the three timepoints in the RGB color channels. In the resulting maps, blue indicates new lesions, pink shows tissue changes, and white represents lesion stability. RESULTS Color-coded change maps significantly improved visualization of small brain tumors and metastases compared to standard imaging. The maps revealed clear associations between edema progression and lesion development in corresponding regions. Most importantly, they enhanced delineation of tumor boundaries that were not apparent in conventional T1 post-contrast images. Clinical Implications: This method has the potential to improve tumor resection accuracy during gamma knife surgery, thereby reducing recurrence risk. The enhanced boundary visualization could be particularly valuable for surgical planning and treatment monitoring. Future Directions: We are currently processing the complete NYUMets dataset to validate our technique across all cases. Future work will explore using these color-coded change maps as saliency maps in neural networks to better predict treatment response versus cancer progression.