Predicting true age from T1 and resting state data.
Started @ NeuroHackademy 2018
Presentation of preliminary results on validation set: https://docs.google.com/presentation/d/1UoVpXbJ80b4R1JWPdWe5TKYDQd09h3MW30N0h0sugmU/edit?usp=sharing
Split CNP_T1_gridsearch into gridsearch and preprocessing.
- Rerun gridsearch.
- T1 for clinical population.
- Run crossvalidation instead of train-validation split.
- Try to predict clinical population from NKI
- Predict NKI and CNP tuned models
- Maybe train NKI models on CNP clinical data
- Train multimodel stacked random forest models.
- Clean NKI T1 data
Make example of transforming T1 and resting state for nilearn.
1. CNP dataset
The model was trained and validated using the control sample (n = 122) from the Consortium for Neuropsychiatric Phenomics dataset (information about preprocessing of the rs-MRI and T1 scans can be found at https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5664981/). The clinical dataset was comprised of three samples:
- ADHD (n = 40)
- Bipolar (n = 49)
- Schizophrenia (n = 50)
1.1.1. Preprocessing DONE
1.1.2. Gridsearch DONE
1.2. Resting state functional connectivity (func)
1.2.1. Preprocessing DONE
- Parcell-based Pearson's correlation matrix: A parcel-based functional connectivity matrix was obtained from the resting-state functional data of each subject using the the 197- and 444-scale BASC atlases (Bellec et al., 2010). Parcel time-series were obtained by averaging the voxel-specific time-series within each parcel. A functional connectivity matrix was obtained for each subject using Pearson's correlation between the derived parcel time-series. Subsequently, the upper triangular of the connectivity matrix of each subject was extracted and reshaped into a vector that was used to predict age.
1.2.2. Gridsearch DONE
1.2.3. Validation DONE
2.2.1. See how well each dataset can predict the other (CNK and NKI)
Gorgolewski, K. J., Durnez, J., & Poldrack, R. A. (2017). Preprocessed Consortium for Neuropsychiatric Phenomics dataset. F1000Research, 6.
Liem, F., Varoquaux, G., Kynast, J., Beyer, F., Masouleh, S. K., Huntenburg, J. M., ... & Riedel-Heller, S. (2017). Predicting brain-age from multimodal imaging data captures cognitive impairment. NeuroImage, 148, 179-188.
Nooner, K. B., Colcombe, S., Tobe, R., Mennes, M., Benedict, M., Moreno, A., ... & Sikka, S. (2012). The NKI-Rockland sample: a model for accelerating the pace of discovery science in psychiatry. Frontiers in neuroscience, 6, 152.