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: | Guilherme Pombo |
| Institution: | University College London |
| Department: | High-Dimensional Neurology |
| Country: | |
| Proposed Analysis: | The manifestations of neurological diseases in the imaged brain are complex, reflecting the intersection of pathological, biological and instrumental forms of variation. A signal of interest here must typically be disentangled from a rich, widely distributed collection of interacting factors: some irrelevant, others critical. The revolution in high-dimensional modelling ushered by deep learning has given rise to promising applications in the realm of brain imaging. Given sufficiently informative data, through end-to-end training, a deep neural network model can implicitly find a decomposition of the image that best supports the task it is deployed to solve. Crucially, the model must rely on the data to distinguish between the target, foreground signal, and the incidental, background context in which it is embedded. Simple geometric context independence, such as translation invariance, is readily achievable through standard augmentations like rotation and translation. But where the context is itself complex -- for example, brain age -- no simple remedy is available. Retained sensitivity to context here not only impairs fidelity, it introduces vulnerability to distributional shifts, and may inject bias through irrelevant natural (confounder) or sampling (collider) correlations. The class imbalance and inadequate data representation common in the clinical domain can only amplify the risks. - I am currently a PhD student investigating how Deep Learning often learns unwanted biases from brain imaging data and how to ameliorate this phenomena. I believe the ADNI dataset could be a great asset in my research to obtain Deep learning models which are fairer (achieve similar performance) across all demographics. |
| Additional Investigators |
