基于异构医学影像数据的深度挖掘技术及中枢神经系统重大疾病的精准预测

结题报告

项目介绍

AI项目解读

基本信息

批准号：
61672236
项目类别：
面上项目
资助金额：
64.0万
负责人：
王骏
依托单位：
华东师范大学
学科分类：
F06.人工智能
结题年份：
2020
批准年份：
2016
项目状态：
已结题
起止时间：
2017-01-01 至2020-12-31

项目参与者：
李郁欣；浦剑；王祥丰；鲍奕仿；纪文迪；王丹萌；邵欣蔚；张翱；张文杰；
关键词：
Distributed Tumor Fusion Deep Large-scale Optimization Metric Learning Brain Multi-modal

项目摘要

With the development of medical imaging equipment, the number of heterogeneous multi-modal images grows rapidly. How to take advantage of the computer power to reveal some useful information is the essential part of precision medicine and individualized treatment. This project will focus on the medical imaging data (e.g., MRI, PET, and CT) to perform accurate prediction for brain tumors, using a uniform depth learning architecture. On the basis of the existing research work, this project will design new deep learning structures for multi-task and multi-modal learning, including: deep transfer neural networks, multi-modal constrained optimization, multi-task feedback optimization, training and testing with missing modal data, and metric learning and fast indexing with multi-modal data. Finally, we are able to perform automatic analysis of medical imaging data, processing, and accurate prediction, no matter based on single mode or multi-modal data. Research results of this project can both enrich the basic theory of machine learning and artificial intelligence, but also provide fast and accurate information support for the medical radiologists. This project has a wealth of academic research and cutting-edge innovation, and important practical value as well.

随着数字化医学影像设备的不断进步和丰富，异构多模态影像数据量快速增长，如何借助计算机的力量深度挖掘其中的有用信息是目前以个体化治疗为目标的精准医疗的难点和关键问题。本项目利用统一的深度学习框架，针对危害性和发病率不断提升的脑肿瘤疾病，深度挖掘以MRI，PET和CT等为代表的医疗影像数据，实现对于脑肿瘤疾病的精准预测。在已有研究工作的基础上，设计适用于多模态和多任务学习的深度神经网络，具体包括深度神经网络迁移算法，多模态约束优化，多任务反馈学习，不完整模态数据的训练和测试，和多模态数据的相似度学习及快速索引，最终实现针对单一模态或多模态的医学影像数据自动分析，处理和精准预测。研究成果既可以丰富机器学习和人工智能的基础理论，又可以为临床医生的影像学诊断提供快速准确的信息技术的支撑。本项目研究具有丰富的学术前沿性和创新性，具备非常重要的实际应用价值。