ClarusC64/clinical-neurologic-deterioration-instability-v0.1

该数据集用于评估模型是否能够从简短的临床代理轨迹中检测出神经功能恶化。每个数据行代表三个时间点的简化神经监测场景，任务是根据多个轨迹变量（如GCS轨迹、瞳孔反应轨迹、颅内压代理轨迹、MAP轨迹、脑灌注代理、自动调节代理、渗透疗法反应和干预延迟）判断系统是否保持神经稳定或趋向恶化。数据集还包括干扰变量（如监测噪声和图表噪声），这些变量看似有意义但不单独决定标签。预测目标为：label = 1表示神经功能恶化不稳定，label = 0表示稳定或受控的神经轨迹。数据集作为Clarus稳定性推理基准的一部分，支持神经功能恶化检测、脑灌注不稳定、自动调节失败、基于轨迹的临床推理、潜在稳定性几何和跨域不稳定基准的研究。

This dataset evaluates whether models can detect neurologic deterioration from short clinical proxy trajectories. Each row represents a simplified neurologic monitoring scenario across three time points. The task is to determine whether the system remains neurologically stable or is moving toward deterioration based on multiple trajectory variables (e.g., GCS trajectory, pupil reactivity trajectory, ICP proxy trajectory, MAP trajectory, cerebral perfusion proxy, autoregulation proxy, osmotherapy response, and intervention delay). The dataset also includes decoy variables (e.g., monitor_noise and chart_noise) that appear meaningful but do not determine the label alone. The prediction target is: label = 1 → neurologic deterioration instability, label = 0 → stable or controlled neurologic trajectory. As part of the Clarus Stability Reasoning Benchmark, this dataset supports research into neurologic deterioration detection, cerebral perfusion instability, autoregulation failure, trajectory-based clinical reasoning, latent stability geometry, and cross-domain instability benchmarks.