Researchers have developed a modular agent that can reliably verify spatial relations in CT scans. The system works by breaking down the task into three stages: parsing natural-language queries, localizing anatomical structures, and using geometric rules to make a final decision. This approach outperforms end-to-end vision-language models on a benchmark dataset, achieving an accuracy of 94.1%. The modular design allows for interpretable intermediate representations and auditable reasoning stages. --- Why it matters: This matters because reliable spatial understanding is crucial for medical vision-language systems that aim to support radiological report generation and structured image understanding. Current end-to-end models are weak in controlled spatial reasoning, which can lead to diagnostic accuracy issues. Source: https://arxiv.org/abs/2608.21140