[论文解读] A novel and reliable deep learning web-based tool to detect COVID-19 infection form chest CT-scan

Abdolkarim Saeedi, Maryam Saeedi+1 more|arXiv (Cornell University)|Jun 24, 2020

COVID-19 diagnosis using AI参考文献 5被引用 13

一句话总结

本论文提出了一种基于网络的可靠深度学习工具，通过结合DenseNet进行特征提取和Nu-SVM进行分类的混合模型，实现从胸部CT扫描中检测COVID-19。该系统在单张图像上实现90.61%的准确率、90.80%的召回率、89.76%的精确率以及95.05%的AUC，推理时间仅为每张图像39毫秒。

ABSTRACT

The corona virus is already spread around the world in many countries, and it has taken many lives. Furthermore, the world health organization (WHO) has announced that COVID-19 has reached the global epidemic stage. Early and reliable diagnosis using chest CT-scan can assist medical specialists in vital circumstances. In this work, we introduce a computer aided diagnosis (CAD) web service to detect COVID- 19 online. One of the largest public chest CT-scan databases, containing 746 participants was used in this experiment. A number of well-known deep neural network architectures consisting of ResNet, Inception and MobileNet were inspected to find the most efficient model for the hybrid system. A combination of the Densely connected convolutional network (DenseNet) in order to reduce image dimensions and Nu-SVM as an anti-overfitting bottleneck was chosen to distinguish between COVID-19 and healthy controls. The proposed methodology achieved 90.80% recall, 89.76% precision and 90.61% accuracy. The method also yields an AUC of 95.05%. Ultimately a flask web service is made public through ngrok using the trained models to provide a RESTful COVID-19 detector, which takes only 39 milliseconds to process one image. The source code is also available at this https URL. Based on the findings, it can be inferred that it is feasible to use the proposed technique as an automated tool for diagnosis of COVID-19.

研究动机与目标

• 开发一种自动化、可靠且可访问的基于网络的工具，用于通过胸部CT扫描图像实现COVID-19的早期检测。
• 通过利用深度学习技术，应对全球COVID-19大流行期间及时且准确诊断的挑战。
• 通过减少对放射科医生手动解读CT扫描的依赖，提升医疗专业人员的诊断效率。
• 创建一个公开可用、低延迟的RESTful网络服务，实现实时的COVID-19检测。

提出的方法

• 开发了一种混合深度学习模型，利用DenseNet从胸部CT扫描中提取分层特征并降低图像维度。
• 采用Nu-SVM作为分类器，起到防止过拟合的瓶颈作用，提升模型的泛化能力。
• 在包含746名参与者的公开数据集上对模型进行训练和验证，涵盖COVID-19阳性病例和健康对照组。
• 评估了多种深度神经网络（ResNet、Inception、MobileNet），以选择最适合该混合系统的架构。
• 基于Flask部署了网络服务，使用ngrok托管训练好的模型，通过REST API实现公众访问。
• 系统仅用39毫秒即可处理单张CT图像，确保临床环境中的实时性能。

实验结果

研究问题

• RQ1 基于深度学习的网络服务是否能在从胸部CT扫描中检测COVID-19时实现高准确率和低延迟？
• RQ2 在该诊断任务中，哪种深度神经网络架构与Nu-SVM分类器结合时表现最佳？
• RQ3 是否可行部署一种可靠、公开可访问的基于网络的CAD系统，用于通过CT影像检测COVID-19？
• RQ4 该混合DenseNet-Nu-SVM模型在保持高诊断性能的同时，是否能有效减少过拟合？

主要发现

• 所提出的基于网络的CAD系统在测试集上区分COVID-19与健康对照组的准确率达到90.61%。
• 该模型表现出90.80%的高召回率，表明对阳性病例具有强敏感性。
• 精确率达到89.76%，表明正样本预测可靠，误报率较低。
• 受试者工作特征曲线下面积（AUC）达到95.05%，表明整体诊断区分能力优异。
• 系统仅用39毫秒即可处理单张CT图像，支持在临床环境中实现实时部署。
• 源代码和训练好的模型公开提供，支持透明度与可复现性。