Speaker:Huitong Ding Date:2017.05.31
Supervisors: Ning An, Jiaoyun Yang, Xia Que
Students: Huitong Ding, Xiaohui Yao, Jinge Yan, Yuting Wang, Yue Yin, Chen Tang, You Duan, Jie Liu, Bo Jing, Yue Teng, Hong Ming, Peng Han, Shuo Liu
Personal leave: Siyuan Jiang, Xu Chen
In this session, we explained how to use machine learning techniques, especially Bayesian network, to understand the relationships among heterogeneous multimodal biomarkers.
This paper has two main innovation points. 1. establish a system-level model that can better describe the interactions among biomarkers 2.provide superior diagnostic and prognostic information.
Considerable research efforts have been devoted to establish a disease model of AD that
could lead to greater understanding of the events that occur in AD. Inspired by that, while many details in the disease model are still unknown, we could devoted ourselves to developing models to gain better and more accurate knowledge of the mechanisms of AD pathogenesis and progression to facilitate a range of clinical tasks such as early diagnosis, treatment efficacy evaluation, treatment planning, better clinical trial design, and drug developments.
The main categories of biomarkers are demographic, genotyping, fluid biomarkers, neuroimaging, neuropsychometric test and clinical assessments.
The description about biomarkers analyzed in this study is listed in table below.
This paper don’t show the reason why choose these biomarkers. There are lots of things to do in the feature selection phase.
It also gives a comprehensive explaination about the drawbacks of black-box prediction models, especially regardless of their heterogeneous clinical nature.
In conclusion, this paper use the mixed type Bayesian network to model the interactions among heterogeneous multimodal biomarkers and conduct this study using
ADNI baseline dataset and find that the learned BN model provides findings that are consistent with the AD literature. Both methods could lead to discovery of
different types of relationships among variables, and different applications may require different approaches for optimal performance. Thus, we believe it will also be a
future research direction.