您现在的位置:网站首页>通知公告

通知公告

【医学部学术讲座】——Deep Neural Networks for Automated Prostate Cancer Detection and Diagnosis in Multi-parametric MRI

文章来源: 作者: 发布时间:2018年01月02日 点击数: 字体:

主讲嘉宾: 杨欣 副教授

时间: 2018年1月4日上午 10:00 – 11:30

地点: 深圳大学西丽校区A2-517

主持人: 倪东 教授

主讲嘉宾简介:

杨欣, 2013年3月获加州大学圣塔芭芭拉学校博士学位,2013年4月至2014年7月在美国加州大学圣塔芭芭拉分校从事博士后研究工作,2014年8月至今申请人担任华中科技大学电信学院副教授。杨欣主要从事医学图像分析及辅助诊断等相关领域的研究工作,在相关领域发表高质量期刊会议论文40余篇,其中近三年以第一作者和通信作者发表JCR二区以上论文8篇包括TPAMI, TMI, Medical Image Analysis, TVCG, 及多篇CCF A类会议ACM Multimedia论文和医学领域顶级会议MICCAI论文,合作出版英文著作2部,授权美国发明专利2项,中国软件著作权4项。杨欣2017年入选武汉3551长期创新人才,多次担任CCF A类会议ACM Multimedia Area Track主席,华中科技大学学术前沿青年团队成员。

报告简介(Abstract):Multi-parameter magnetic resonance imaging (mp-MRI) is increasingly popular for prostate cancer (PCa) detection and diagnosis. However, interpreting mp-MRI data which typically contains multiple unregistered 3D sequences, e.g. apparent diffusion coefficient (ADC) and T2-weighted (T2w) images, is time-consuming and demands special expertise, limiting its usage for large-scale PCa screening. Therefore, solutions to computer-aided detection and diagnosis of PCa in mp-MRI images are highly desirable. Most recent advances in automated methods for PCa detection employ several separate steps, including multimodal image registration, prostate segmentation, voxel-level classification for candidate generation and a region-level classification for verification. Features used in each classification stage are handcrafted. In addition, each step is optimized individually without considering the error tolerance of other steps. As a result, they could either involve unnecessary computational cost or suffer from errors accumulated over steps. In this talk we will introduce a series of our recent works on utilizing deep convolutional neural networks (CNN) for automated PCa detection and diagnosis. We will introduce our co-trained weakly-supervised CNNs which can concurrently identify the presence of PCa in an image and localize lesions. Our weakly-supervised CNNs can learn representative lesion features from entire prostate images with only image-level labels indicating the presence or absence of PCa, significantly alleviating the manual annotation efforts in clinical usage. Multi-model information from ADC and T2w are fused implicitly in CNNs so that the feature learning process of each modality can be mutually guided by each other to capture highly representative PCa-relevant features. We will also introduce our Tissue Deformation Network (TDN) for automated prostate detection and multimodal registration. The TDN can be directly concatenated with our weakly-supervised PCa detection CNNs so that all parameters of the entire network can be jointly optimized in an end-to-end manner. Comprehensive evaluation on 360 patient data demonstrates that our system achieves a high accuracy for CS PCa detection and is outperforms the state-of-the-art CNN-based methods and 6-core systematic prostate biopsies.

【打印文章】 【添加收藏】