谢修娟　顾兵

摘   要：细菌耐药性的日益加剧，以及目前的耐药性检测方法周期长等问题，给临床第一时间准确用药带来极大的挑战和困境. 为此，本文将探索深度学习技术在抗菌药物耐药性预测中的应用，提出一种融合注意力机制的双通道卷积神经网络模型，通过上下两个通道对建模后的送检数据做不同粒度的特征提取，每个通道经过卷积和池化后引入注意力机制，聚焦重要的特征信息，而后将两个通道的特征进行融合，从而完成分类输出. 将模型在某三甲医院细菌药敏检测历史数据集上，与多种不同方法进行对比实验，结果表明，本文所提出方法在分类准确度F值指标中平均实现20.35%的提升，同时在小样本分类上表现出更好的效果.

关键词：深度学习;卷积神经网络;注意力机制;耐药性预测

中图分类号：TP301                                   文献标志码：A

Research on Antimicrobial Resistance Analysis Based on Deep Learning

XIE Xiujuan GU Bing

（1. Department of Computer Engineering，Southeast University Chengxian College，Nanjing 210000，China;

2. College of Medical Technology，Xuzhou Medical University，Xuzhou 221004，China）

Abstract：The increasing drug resistance of bacteria，as well as the long cycle of current drug resistance testing methods，bring great challenges and difficulties to accurate drug use at the first time in clinic.Therefore，this paper will explore the application of deep learning technology in the prediction of antimicrobial resistance，and proposes a dual-channel convolution neural network model integrating attention mechanisms. Through the upper and lower channels，different granularity features are extracted from laboratory data after modeling. After convolution and pooling，an attention mechanism is introduced in each channel to focus on important feature information，and then the features of the two channels are fused to complete the classification output. The model is applied to the historical data set of bacterial drug sensitivity test in a tertiary hospital，and compared with other methods.The results show that the proposed method achieves an average improvement of 20.35% in F-value index of classification accuracy，and performs better in small sample classification.

Key words：deep learning;convolutional neural network;attention mechanism;drug resistance prediction

在互聯网技术高速发展的今天，医疗、教育、金融、电商等各个行业的数据以“爆炸”式速度在增长，如何从海量数据中，提取有价值的信息变得越来越重要. 深度学习作为大数据分析的关键技术之一[1]，它将特征提取与分类器有机融合，能从大量输入信息中自动学习特征并完成分类，在图像识别、语音识别、文本分类及情感分析等诸多领域取得了较好的实践效果[2-3]. 近年来，不少学者开始探索医疗卫生行业大数据的智能化处理，尝试运用深度学习技术处理医学大数据，主要集中于对医学图像、电子病历、基因……