基于深度学习的活体细胞有丝分裂检测方法  

Deep Learning-Based Detection Method for Mitosis in Living Cells

作　　者：柯宝生 李颖[1,2,3] 任振波 邸江磊[1,2,3] 赵建林 Ke Baosheng;Li Ying;Ren Zhenbo;Di Jianglei;Zhao Jianlin(School of Physical Science and Technology,Northwestern Polytechnical University,Xi'an,Shaanxi 710129,China;Shaanxi Key Laboratory of Optical Information Technology,Xi'an,Shaanxi 710129,China;Key Laboratory of Material Physics and Chemistry Under Extraordinary Conditions,Ministry of Education,Xi'an,Shaanxi 710129,China)

机构地区：[1]西北工业大学物理科学与技术学院,陕西西安710129 [2]陕西省光信息技术重点实验室,陕西西安710129 [3]超常条件材料物理与化学教育部重点实验室,陕西西安710129

出　　处：《光学学报》2021年第15期88-97,共10页Acta Optica Sinica

基　　金：国家自然科学基金(61927810,62075183,61905197);中央高校基本科研业务费(310201911qd002)。

摘　　要：活体细胞有丝分裂过程的发生具有时间和空间上的随机性,自动识别并准确定位活体细胞的有丝分裂对科研人员而言充满挑战。提出一种基于深度学习的自动识别并定位活体细胞有丝分裂的检测方法。通过改进YOLOv3主干网络并引入注意力机制,构建名为DetectNet的深度神经网络。在明场显微成像条件下,获取多尺寸活体细胞图像并构建数据集对网络进行训练,并对DetectNet与多个目标检测算法进行对比,验证其有效性。实验结果表明,针对活体细胞的明场显微图像,DetectNet能够高效地从不同尺寸大视场图像中直接识别并定位有丝分裂细胞,同时具有较高的检测精度和较快的检测速度,因而在生物和医学领域具有非常大的潜在应用价值。Owing to the spatiotemporal randomness of mitosis, the automatic identification and accurate location of mitosis in living cells are challenging tasks for researchers. Herein, a deep learning-based detection method was proposed to automatically identify and locate mitosis in living cells. Here, we built a deep neural network called DetectNet by improving the backbone network of YOLOv3 and introducing an attention mechanism. Under the condition of bright-field microscopic imaging, multiscale images of living cells were acquired and then a dataset was constructed to train the network. The trained network DetectNet was compared with multiple object detection algorithms, and its effectiveness was verified. Experimental results show that aiming at the bright-field microscopic images, DetectNet can directly identify and locate mitosis from the multiscale live cell images with a large field, achieving a higher detection accuracy and faster detection speed compared with other multiple object detection algorithms. Thus, DetectNet shows a great potential application value in the fields of biology and medicine.

关 键 词：成像系统 活体细胞 有丝分裂 深度学习 目标检测算法 明场显微成像 

分 类 号：O436.1[机械工程—光学工程]