基于多模态融合的儿童慢性咳嗽人工智能辅助诊断技术研究

慢性咳嗽是儿科最常见的临床症状之一，其病因复杂又与成人不尽相同，是儿科呼吸道疾病诊断的重点和难点。项目拟在课题组临床病例和结构化数据辅助诊断研究成果基础上，从人工智能和医学交叉角度，通过引入多模态数据融合方法，提出一套面向医生和病人的儿童慢性咳嗽辅助诊断理论方法，解决儿童慢性咳嗽细粒度疾病计算机自动识别这一难点问题，以增强儿科慢性咳嗽高水平供给能力。项目围绕细粒度识别、小样本建模和多模态数据融合三个难点问题，提出面向慢性咳嗽的数据建模与分析方法，提出多尺度类别树、特征嵌入与融合、小样本类别数据增强、长尾分布训练等多种适合儿童慢性咳嗽精准识别的人工智能创新算法。项目拟研制结果可视化、算法仿真验证平台，构建10万以上病历样本的儿童慢性咳嗽相关呼吸道疾病数据库，并在医院实际部署项目成果，探索辅助诊断系统在儿科医院落地的实践解决方案。

Chronic cough in children is one of the most common clinical symptoms in pediatrics, and is drawing more and more attention because of its complicated etiology (which is also different from adults) and difficulty in diagnosis. From the perspective of artificial intelligence and medicine, this project proposes a theoretical system of AI-assisted chronic cough diagnostic system by introducing multi-modal data fusion methods, so as to solve the problem of recognizing fine-grained diseases of chronic cough in children. This project conducts research by focusing on fine-grained recognition, few-shot modeling and multi-modal data fusion. We study the data modeling and analysis methods of chronic cough, and put forward to AI-based chronic cough recognition methods and corresponding evaluation methods. Furthermore, we propose some AI modules of multi-scale category tree, feature infusion and fusion model, few-shot data augmentation, long-tail training, and try to solve the analysis, modeling, recognition and deployment problems of chronic cough diagnostic system. In addition, we verify the model and methods through developing prediction result visualization platform and simulated deployment platform. A dataset of over 100,000 chronic cough case samples will be collected to evaluate the proposed methods, and the system will be deployed in at least 1 hospital to explore the practical solution of diagnostic system.