Brain metastasis detection improves with combined 3D deep learning views
Scale-Aware 3D Deep Learning for Robust Brain Metastasis Detection in Multimodal MRI
Computer Vision and Pattern RecognitionMachine Learning
Summary
Finding brain metastases in MRI scans is tough because these cancer spots come in many sizes, and some are very tiny. The authors studied how looking at MRI images from different zoom levels and combining those views can help detect these spots better. They trained two separate 3D deep learning models that look at the brain in different sized blocks and then merged their results. This combined approach found more true cancer spots and made fewer mistakes compared to using just one model.
What this means in practice
- •For medical imaging software developers: Develop software for brain MRI analysis that improves detection accuracy by fusing outputs from models analyzing different image scales.
- •For radiology workflow teams: Integrate scale-aware deep learning tools into MRI reading pipelines to reduce false alarms and improve confidence in detecting brain metastases.
Authors
Sylvain Jaume, Hongming Wang, Simon K. Warfield
Abstract
Detecting brain metastases in magnetic resonance imaging (MRI) remains challenging because lesions vary widely in size and appearance, with very small metastases occupying only a minute fraction of a three-dimensional input. We investigate whether combining different spatial fields of view (FOVs) improves lesion detection in multimodal MRI and present a scale-aware 3D deep-learning framework. The method uses independently trained $96^3$ and $64^3$ 3D U-Nets whose whole-volume probability maps are combined by weighted late fusion. This design allows us to study the effect of spatial context separately from image resolution and modality choice. On a 97-patient development cohort, cross-FOV fusion improved lesion-level precision and F1 while substantially reducing false positives relative to the individual models. A same-FOV ensemble control showed that these gains were not explained solely by averaging independently trained networks, supporting a contribution from complementary spatial context. An exploratory cross-FOV agreement filter reduced false positives but did not improve overall F1. These results support cross-FOV probability fusion as a simple and computationally practical strategy for improving the precision-false-positive trade-off in 3D brain-metastasis detection.