Search papers, labs, and topics across Lattice.
The paper introduces CARE-X, a chest X-ray Vision-Language Model (VLM) that integrates auxiliary discriminative supervision with reward-aligned learning to enhance clinical utility. By employing focal-loss classification and composite-loss grounding alongside traditional language modeling, CARE-X achieves superior diagnostic predictions, spatial localization, and report quality. The model demonstrates state-of-the-art performance across multiple benchmarks, significantly improving visual question answering accuracy and measurement-dependent diagnoses through hybrid inference techniques.
CARE-X achieves a remarkable 94.0% accuracy in visual question answering, outperforming existing models by 6 percentage points while also enhancing report quality and spatial localization.
A clinically useful chest X-ray system must go beyond fluent report generation: it should classify findings with tunable decision thresholds, localize them spatially, and derive the anatomical measurements upon which many diagnoses depend. Today's Vision-Language Models (VLMs) treat these as separate problems, if they address them at all, leaving a gap between what radiologists need and what generative models provide. We introduce CARE-X, a chest X-ray VLM that narrows this gap by unifying auxiliary discriminative supervision with reward-aligned generation. CARE-X augments its generative backbone with focal-loss classification and composite-loss grounding heads, co-trained alongside the language-modeling objective. This auxiliary supervision produces discriminative diagnostic predictions with tunable decision thresholds and precise spatial localization while also improving report quality, providing evidence that structured prediction and generation reinforce one another. Building on this foundation, Decoupled Clip and Dynamic Sampling Policy Optimization (DAPO) leverages task-specific reward signals for report generation, visual question answering (VQA), and spatial grounding, directly optimizing the clinical quality metrics that matter in practice. The result is state-of-the-art performance on the majority of metrics across four report-generation benchmarks, 94.0% VQA accuracy on ReXVQA (+6.0 pp over the next-best baseline), and generative spatial decoding that reaches near parity with dedicated detection heads. Separately, to address measurement-dependent diagnoses, we couple Qwen3-VL-4B-Instruct with native tool-calling capabilities for invoking deterministic measurement tools, while retaining full visual access to the image. This hybrid inference yields +43.6 pp average F1 over perception-only baselines across five measurement-dependent conditions.
