MediRound: Multi-round segmentation of entities in medical images

Medical image segmentation has advanced significantly with the integration of language models, but most current solutions are limited to single-round dialogues, making them difficult to use in educational environments where progressive reasoning is required. In this context, MediRound emerges, a proposal that introduces multi-round entity segmentation in medical images (MEMR-Seg). This approach allows generating segmentation masks through iterative queries with entity-level reasoning, enabling students to develop their anatomical and pathological understanding step by step. To train and evaluate the model, the MR-MedSeg dataset was built, with over 177,000 multi-round segmentation dialogues, and a lightweight judgment and correction mechanism was designed to mitigate the typical error propagation of chained pipelines. The implementation of this type of AI for businesses demonstrates how artificial intelligence can transform medical training and assisted diagnosis. At Q2BSTUDIO, we develop custom applications that integrate AI models, AWS and Azure cloud services to scale infrastructures, and AI agents that automate complex processes. Additionally, we offer Power BI to visualize results and cybersecurity to protect sensitive data. These capabilities allow solutions like MediRound to be brought into production environments, combining business intelligence services and custom software to meet the specific needs of the healthcare sector.