Please enable JavaScript.

Coggle requires JavaScript to display documents.

Alzheimer's Disease (AD), Note: Instead of synthetic sampling can…

- - - - Loss functions
      - Random sampling
      - Cosine distance
- - - - Classified normal, mild cognitive impairment, and AD images
      - 85.7% accuracy
    - - Used AD Neuroimaging Initiative Dataset (ADNI)
      - 89.% accuracy AD vs Normal images
    - - 3D images
      - ADNI Database
      - 97% accuracy
    - - DenseNets variations
      - OASIS Dataset
      - 95.23% accuracy
    - - ADNI Dataset
      - 95% accuracy
    - - Automated technique
      - 73.75% accuracy
  - - - Low generalizability
- - - - Visualize high dimension data
      - Reduce original data to 2 or 3 dimensions
    - - Compresses inputs
      - Creates latent space
      - Samples representations from latent space to reconstruct input
    - - Combination of VAE and Generative Adversarial Networks
      - Best method
  - - - Very mild demented (VMD)
      - mild demented (MD)
      - Non-demented (ND)
      - Moderated demented (MDTD)
    - - Up-sampling is risky
        
        Will lead to extremely high or low class variations
      - Pre-process instead
        
        Standard Gaussian noise
        
        Image saturation
        
        Random right and left flip
        
        Slight rotation
        
        Center crop
        
        Use of image entropy analysis
  - - - Based on embedding size after dimension reduction
      - Use of cosine similarity measure
    - - Sum of two loss functions
      - Unsupervised loss function
      - Supervised loss function
- - - - Model with similarity graph
      - Model WITHOUT similarity graph
      - Comparison of data representation
      - Comparison of clustering techniques
        
        K-means
        
        Hierarchical clustering
      - Similarity with AAE + K0means achieved the best performance
    - - Effect of regularization
        
        CNN + Similarity graph outperformed baseline models
      - Outperforms other state-of-the-art techniques for AD detection