基于深度学习与阈值分割的线虫识别-《徐州医科大学学报》

文章信息/Info

Title:: Application of machine learning and threshold segmentation in nematode recognition

作者:: 杜昕泽1; 金治宇2; 施朋3; 赵鱼良1; 俸浩然1; 张璇1; 朱作斌4*; 1. 徐州医科大学第一临床医学院，江苏徐州221004；2. 南京大学电子科学与工程学院，江苏南京210020；3. 徐州医科大学麻醉学院；4. 徐州医科大学遗传学教研室,徐州医科大学遗传基础与临床应用重点实验室,徐州市医学遗传与转化工程研究中心

摘要:: 目的基于深度学习和阈值分割的方法实现对图像中秀丽隐杆线虫数量和聚集程度的统计，提高行为实验的效率和准确度，便于利用线虫等小型模式生物进行高通量筛选针对神经行为的潜在药物。方法采集约5000张含有线虫的培养基样本，基于深度学习与阈值分割的方法设计线虫识别检测算法，实现对培养基中线虫的自动计数与聚集程度判断。结果本研究基于深度学习和阈值分割算法，设计了自动统计线虫数量及聚集程度的软件，该软件与人工统计结果比较，分析时间显著低于人工计数，对每张图像的检测时间在毫秒级别，且检测的准确率较好，针对单只线虫的检测准确率超过90%。结论本方法解决了人工统计图像中小型模式生物秀丽隐杆线虫总数和聚集程度耗时长、效率低等问题，借助本研究设计软件，可以高效的筛选作用于神经行为相关药物。

Abstract:: ob＜x＞jective ba＜x＞sed on machine learning and threshold segmentation, the number and aggregation degree of Caenorhabditis elegans in images can be counted, and the efficiency and accuracy of social behavior experiments can be improved, to facilitate the use of small model organisms such as nematodes for high-throughput screening of potential neurobehavioral drugs。Methods About 5000 samples of medium containing nematodes were collected, and an algorithm of nematode identification and detection ba＜x＞sed on machine learning and threshold segmentation was designed。Results ba＜x＞sed on machine learning and threshold segmentation algorithm, a software was designed to automatically count the number and aggregation degree of nematodes. Compared with the results of artificial statistics, the analysis time was significantly lower than that of artificial counting, the detection time of each image is in the millisecond level, and the detection accuracy is better than 90% for single nematode 。Conclusion This method solves the problems of time-consuming and low efficiency of the total number and aggregation of Caenorhabditis elegans, which is a small and medium-sized model organism。