TASKS.md

TO-DOs/Tasks:

Machine Learning Classification (Realistic :D)

1. [✅] Use masks to normalize brain data
2. [✅] Find number of GM, WM, & CSF pixels for each participant
   • Voxel Brightness values: CSF = 1, GM = 2, WM = 3
3. [✅] Bin labels/ages
   • try to get smallest bins possible with good classification
   • [✅] Train-test-split
4. [✅] Feed into ML classifier (i.e., LR) and label with ages
   • Find relationship between GM, WM, & CSF with age
   • See what accuracy we can get predicting age with amount of GM, WM, & CSF

CNN: (NOW A STRETCH GOAL!)

1. [✅] Use masks to normalize brain data
2. [✅] Find Ground Truth Data
   • Voxel Brightness values: CSF = 1, GM = 2, WM = 3
3. Train-test split
   • Randomly split participants: 520 training, ~104 validation & hypertuning, ~132 testing
     • Control for age and sex across splits ()
4. Implement Model
   • 3 layers (fully-connected), 2 hidden layers: ~30 & ~20 neurons each (w/ ReLU)
   • Output layer trained with softmax given pixels from segs_refs
5. Evaluate CNN performance with Dice Score (dice_coef)?
   • 
6. Use Statistics and/or machine learning to identify relationship with participant age
   • Correlation, (non)-linear reggression, mutual formation, etc.