Many of the cellular events that underlie neurodegenerative disease are best captured by continuously imaging live neurons over time. While the advent of robot-assisted microscopy has helped scale such longitudinal live microscopy to high-throughput regimes with the statistical power to detect transient events, time-intensive human annotation is still relied on for analyzing these experiments. We address this fundamental limitation of live microscopy with biomarker-optimized convolutional neural networks (BO-CNN): interpretable computer vision models trained directly on biosensor activity. We demonstrate the ability of BO-CNNs to detect cell death, a fundamental biological process that is typically measured by trained annotators. BO-CNNs detect cell death with super-human accuracy and speed by learning to identify important subcellular morphology associated with cell vitality, despite receiving no explicit supervision to rely on these features. Importantly, these models also uncover novel intranuclear morphology signal that is difficult to spot by eye and has not yet been linked to cell death, but reliably indicates death. BO-CNNs are broadly useful for analyzing any live microscopy and essential for interpreting high-throughput robotic-aided experiments.

中文翻译：

---

使用生物标志物优化的神经网络进行超级人类细胞死亡检测

许多作为神经退行性疾病基础的细胞事件最好通过随着时间的推移对活神经元进行连续成像来捕捉。虽然机器人辅助显微镜的出现帮助将这种纵向实时显微镜扩展到具有检测瞬态事件的统计能力的高通量方案，但仍然依赖于时间密集的人工注释来分析这些实验。我们使用生物标志物优化的卷积神经网络 (BO-CNN) 解决了实时显微镜的这一基本限制：直接在生物传感器活动上训练的可解释计算机视觉模型。我们展示了 BO-CNN 检测细胞死亡的能力，这是一个基本的生物过程，通常由训练有素的注释者测量。BO-CNNs 通过学习识别与细胞活力相关的重要亚细胞形态，以超人类的准确性和速度检测细胞死亡，尽管没有接受依赖这些特征的明确监督。重要的是，这些模型还发现了新的核内形态信号，这些信号很难被肉眼发现，并且尚未与细胞死亡相关联，但可以可靠地表明死亡。BO-CNN 可广泛用于分析任何实时显微镜，并且对于解释高通量机器人辅助实验至关重要。