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: | Nico Scherf |
| Institution: | Max Planck Institute for Human Cognitive and Brain Sciences |
| Department: | Neural Data Science and Statistical Computing |
| Country: | |
| Proposed Analysis: | Noninvasive MRI brain imaging can reveal atrophy patterns in neurodegenerative diseases that can serve as critical morphological biomarkers for diagnosis. While focal atrophy patterns are relatively specific for some diseases, the general patterns are less conclusive requiring specialized neuroradiologists to detect and interpret the patterns correctly. Machine Learning tools would open a new way to analyze these complex relationships in high dimensional data more reliably. In this work, we will focus on the multi-class prediction from MRI data using Deep Neural Networks as the key for building computational aid systems for differential diagnosis. While Deep Convolutional Neural Networks (CNN) were tremendously successful in solving a lot of computational vision tasks, in biomedical imaging, they tend to overfit to the specifics of the underlying imaging modality. The overfitting effect is particularly problematic for analyzing clinical, non-quantitative MRI data where the choice of signal weighting depends on the preferences of the radiologists. (As a result, different clinics use different MRI modalities with different intensities and contrasts.) In this work, we will develop a new method for structure-aware classification of neurodegenerative syndromes that is independent of specific MRI contrasts by training the neural networks on different modalities. To achieve this robustness to contrasts, we will design targeted augmentation methods and also assess related methods considered in the literature. Based on this architecture, we will train different classifiers across multiple syndromes and adapt these classifiers to clinical needs by focusing on diseases that are relevant to differential diagnosis. |
| Additional Investigators | |
| Investigator's Name: | Sebastian Niehaus |
| Proposed Analysis: | Noninvasive MRI brain imaging can reveal atrophy patterns in neurodegenerative diseases that can serve as critical morphological biomarkers for diagnosis. While focal atrophy patterns are relatively specific for some diseases, the general patterns are less conclusive requiring specialized neuroradiologists to detect and interpret the patterns correctly. Machine Learning tools would open a new way to analyze these complex relationships in high dimensional data more reliably. In this work, we will focus on the multi-class prediction from MRI data using Deep Neural Networks as the key for building computational aid systems for differential diagnosis. While Deep Convolutional Neural Networks (CNN) were tremendously successful in solving a lot of computational vision tasks, in biomedical imaging, they tend to overfit to the specifics of the underlying imaging modality. The overfitting effect is particularly problematic for analyzing clinical, non-quantitative MRI data where the choice of signal weighting depends on the preferences of the radiologists. (As a result, different clinics use different MRI modalities with different intensities and contrasts.) In this work, we will develop a new method for structure-aware classification of neurodegenerative syndromes that is independent of specific MRI contrasts by training the neural networks on different modalities. To achieve this robustness to contrasts, we will design targeted augmentation methods and also assess related methods considered in the literature. Based on this architecture, we will train different classifiers across multiple syndromes and adapt these classifiers to clinical needs by focusing on diseases that are relevant to differential diagnosis. |
