There are many active research projects accessing and applying shared ADNI data. Use the search above to find specific research focuses on the active ADNI investigations. This information is requested annually as a requirement for data access.
| Principal Investigator | |
| Principal Investigator's Name: | Qingrun Zhang |
| Institution: | University of Calgary |
| Department: | Department of Mathematics and Statistics |
| Country: | |
| Proposed Analysis: | This research work is carried out with the idea of applying computer vision (CV) machine learning techniques to the brain MRI data for Alzheimer’s disease (AD) detection and diagnosis. Two recent technological advances have made this possible. One of them is that large-scale MRI image datasets can be obtained. The other is the rapidly evolving machine learning algorithm, especially the CNN-based CV algorithm. Applying machine learning algorithms to the processing and interpretation of medical images is a hot topic in recent research. However, robust and accurate machine learning models are still important for complex tasks. Existing machine learning based CV methods rely on the global feature of the image. However, AD-related features concentrate in different tissue regions, for example as grey matter (GM), white matter (WM), and cerebrospinal fluid (CSF). This feature makes it important to design a model that introduces two CV tasks: image segmentation and classification. This feature is also important and offers the possibility to construct explainable and trustworthy machine learning models for AD detection. The idea of this study is to analyze the connection between the image, disease, and genetic information. Aim 1: Apply traditional machine learning based CV algorithms for brain MRI image data segmentation and pattern recognition. Aim 2: Design robust and accurate machine learning models for AD detection and diagnosis. Aim 3: Study the connection between image-derived phenotypes, the risk of diseases, and genomics data. |
| Additional Investigators | |
| Investigator's Name: | Da Li |
| Proposed Analysis: | This scientific research work contains the following aspects: 1. AD detection from MRI. Existing machine learning based computer vision (CV) methods rely on the global feature of the image. However, AD-related features concentrate in different tissue regions, for example as grey matter (GM), white matter (WM), and cerebrospinal fluid (CSF). This feature makes it is important to design a model that combines two CV tasks: image segmentation and classification. This is also important and offers the possibility for construct explainable and trustworthy machine learning models for AD detection. We will study the above ideas and propose the corresponding machine learning models. 2. Imply the semi-supervised learning technique in brain MRI data. One of the most important issues for medical images is the amount of labeled data is small due to the high costs of labor. To overcome this issue, we will introduce semi-supervised learning with generative models. We will study different generative models base on the ADNI dataset and explore the introduction of generative techniques to improve the generalization of the machine learning model. Also, we will study contrastive learning strategies on MRIs data to see if self-supervised models have the potential to be introduced into AD detection. 3. Image-derived phenotypes (IDPs) (distinct measures of brain structure and function) study for AD disease. 3 possible topics: 1) how to find the IDPs 2) what is the connection between the IDPs and the AD disease. 3) Possible connection with TWAS? |
| Investigator's Name: | Qing Li |
| Proposed Analysis: | Elucidating the genetic basis of brain disorders can help reduce the social-economic burden and improve quality of life. In the field of statistic genetics, one of the ultimate goals is to identify causal genetic variants associated with diseases, and the other one is genotype-based phenotype prediction, which may help tailor precision medication for individuals. However, there are still gaps between statistic prediction and real traits. This is partly due to the high dimensionality of genetic variants together with relatively small samples, which leads to overfitting. Overfitting is a common problem in statistical learning, which is especially detrimental when the models only statistically fit the training data without reflecting genuine biological association. In this project, we aim to bridge the gap by integrating multi-omics data and characterizing statistical models to form powerful predictors. We hypothesize that the genotype and the focal phenotype are linked by internal phenotypes such as various ‘omics and brain features (e.g., hippocampal volume). So instead of using millions of genetic data only, we intend to assimilate biological information from multi-scale ‘omics and brain images into machine learning algorithms such as regularization, kernel machine, Bayesian method, and etc. Specifically, we plan to utilize the-state-of-the-art machine learning technique to extract images features sophistically. In this way, noisy genetic variants are eliminated, and meaningful biological information will stand out. The success of this project will further our understanding of the genetic basis of brain diseases and provide a novel approach to study the pathology of brain diseases. |
| Investigator's Name: | Jiayi Bian |
| Proposed Analysis: | Brain disorders comes in different forms but many of them are caused by genetics basis. The study of brain disorders is way more complex than other diseases. For many years' effort, there are still gaps between statistic prediction and real clinical traits. This is partly due to the relatively small number of samples but a high dimensionality of genetic variants in the medical field, so one easily runs the risk of overfitting. This project will bridge the gap by integrating multi-omics data and characterizing statistical models to form powerful predictors. We hypothesize that the genotype and the external phenotype of organisms are linked by internal phenotypes such as trans-omics and brain features such as hippocampal volume, etc. So instead of using millions of genetic data only, we intend to assimilate biological information from expression data and Magnetic Resonance Imaging (MRI) images data, etc. into machine learning algorithms such as regularization, kernel machine, Bayesian method, etc. to develop novel statistical good fitted models. In this way, noisy genetic variants are eliminated, and meaningful biological information will stand out. |
| Investigator's Name: | Rushani Nilakshika Kahanda Liyanage |
| Proposed Analysis: | Genetic variants, which are divided into common and uncommon variants based on the prevalence of each allele in the population, account for a reasonable share of human disorders. Usually disease causal variants are under negative selection and as a result they have become rare in the population. Due to low frequency of rare variants, there is no use of using standard TWAS for rare- variants. Therefore, novel methods to identify genes should be made using the rare variants.This work extends TWAS protocol to multi-omics data which could first select rare variants that are likely associated with a data-bridge(i.e. brain images,proteins) then use selected variants to conduct aggregation-based test. |
