基于青光眼多源异构医疗数据的深度学习模型与可视化方法研究

In recent years, the rapid development of deep learning methods has brought unprecedented opportunities and challenges for the application of Data Science in the medical field. On the top of the list is to design and optimize the deep learning model for the data that are of large volume and multiple modalities, and are often hard to obtain annotations. This project has taken Glaucoma, a common but complex disease, as the research object. Based on the multi-source heterogeneous medical data, we aim to exploit the deep learning model and visualization methods to achieve a comprehensive understanding of Glaucoma disease. Specifically, the four goals are: 1) combine Transfer Learning , Semi-supervised learning and Graph-cut-Graph-search algorithm to conduct the quantitative analysis based on small sample deep learning; 2) to achieve multi-modality fusion of images on human posterior eye, based on their structural and functional information; 3) design Asynchronous-Input Deep Neural Network (AIDNN) model to achieve joint analysis of the multi-source medical data; 4) to improve the interpretability of deep learning model in clinical application through visualization. This project will provide key methodological support for the intelligent diagnosis of glaucomatous diseases and effectively promote its diagnostic accuracy and therapeutic treatment effect, and ultimately smoothen the application of data science in the medical field.

近年来，深度学习方法的飞速发展为数据科学在医疗领域的应用带来了前所未有的机遇和挑战。如何针对医疗数据体量大、种类多、标注难以获得的特点，来设计和优化深度学习模型，是这一领域迫切需要解决的问题。本项目选取青光眼这种高发常见的致盲性疾病作为研究对象，利用深度学习模型与可视化方法，在多源异构医疗数据的基础上，对青光眼疾病实现全面解读，拟：1）结合迁移学习、半监督式学习与图割图搜索算法，研究基于小样本深度学习的精确定量分析算法；2）基于不同类型眼底图像的结构与功能信息，实现眼底多模态影像的融合；3）利用异步输入深度神经网络分析包括影像、临床等检测数据，实现多源医疗数据的联合分析；4）通过可视化方法，完善临床应用中深度学习模型的可解释性。本项目将为切实做到对青光眼疾病的智能诊断提供关键的方法学支持，以有效地提高青光眼疾病的诊断精度和治疗效果为切入点，有力推动数据科学在医疗领域的顺利发展。

近年来，深度学习方法的飞速发展为数据科学在医疗领域的应用带来了前所未有的机遇和挑战。如何针对医疗数据体量大、种类多、标注难以获得的特点，来设计和优化深度学习模型，是这一领域迫切需要解决的问题。本项目在面向医疗图像处理的方法学研究上，产生一系列研究成果，包括小样本学习、多模态领域迁移、复杂结构的图像分割方法等。在应用方面，本项目针对眼底图像实现了稳定高效的眼底血管定量方法。以上图像分析方法也被应用于其它疾病，验证了这些方法在医疗图像，乃至自然图像上的应用可推广性。本项目以对青光眼数据为研究基点，研究和开发一系列面向医疗图像处理的基础分析方法，并在多种疾病上进行了验证，有潜力推动数据科学在医疗领域的进一步发展。