基于深度学习的自助CT识别平台

Self-Service CT Recognition Platform Based on Deep Learning

传统看病流程有相当一部分的时间都耗费在排队取片、找医生诊断上,这给紧张的医疗资源和紧急的病人情况带来了极大的挑战。快速诊断并进行早期干预成为一种趋势。在此背景下,本项目基于人工智能设计了一个能够自主识别病毒的CT平台,使用理论较为成熟的深度残差网络进行CT识别。卷积神经网络拥有强大的自适应性、学习性、全局最优等功能,在CT识别中表现出较好的性能。用户通过手机扫描二维码上传肺部CT图像至云端服务器,训练好的CT病毒模型对上传的图片按照三个方面即正常、细菌性肺炎和病毒性肺炎进行预估处理,并将预估值以图片的形式返回到用户端,使医生在模型的辅助下能够快速地对就诊人进行诊断,可进一步减少医生在确诊新冠肺炎疑似病例上的工作量。

A considerable part of the traditional medical treatment process is spent queuing up to get pictures and finding doctors for diagnosis, which brings great challenges to the shortage of medical re-sources and urgent patient conditions. Rapid diagnosis and early intervention have become a trend. In this context, this project designs an independent CT virus recognition platform based on artificial intelligence, and uses the deep residual network which is more mature in artificial intelligence to carry out CT recognition. Convolutional neural network has strong self-adaptability, learning and global optimization, and shows good performance in CT recognition. The user uploads lung CT im-ages to the cloud server by scanning two-dimensional code of mobile phone. The trained CT virus model estimates the uploaded images according to three aspects: normal, bacterial pneumonia and viral pneumonia, and returns the estimated values to the user in the form of pictures, so that doc-tors can quickly diagnose patients with the assistance of the model, which can further reduce the workload of doctors in the diagnosis of suspected cases of COVID-19.