快速检测和识别自由行为的秀丽隐杆线虫的全脑成像数据集

我们提出了一种级联解决方案，用于在自由移动的秀丽隐杆线虫中快速识别头节神经元。首先，通过深度学习算法检测一堆荧光图像中的潜在神经区域。其次，将2D神经区域融合成3D神经元实体。第三，通过利用神经元周围的神经元密度分布和神经元之间的相对位置信息，多类人工神经网络将工程化的神经特征向量转化为数字神经元身份。最终总精度可以达到85%以上。

We propose a cascaded solution for rapid identification of head ganglion neurons in freely moving *Caenorhabditis elegans*. First, potential neural regions in a stack of fluorescence microscopy images are detected using deep learning algorithms. Second, 2D neural regions are fused into 3D neuronal entities. Third, a multi-class artificial neural network converts engineered neural feature vectors into digital neuronal identities by leveraging the neuronal density distribution around individual neurons and the relative positional information between neurons. The final overall accuracy can exceed 85%.