Review of machine learning models applied to MRI data for the diagnosis and prognosis of Alzheimer's Disease

  1. MSc thesis
  2. ΧΡΙΣΤΙΝΑ ΜΑΡΙΑ ΒΕΝΕΤΣΙΑΝΟΥ
  3. Βιοπληροφορική και Νευροπληροφορική (ΒΝΠ)
  4. 30 Ιουνίου 2024
  5. Αγγλικά
  6. 63
  7. Χαρίδημος Κονδυλάκης
  8. Alzheimer’s disease, machine learning models, MRI, workflow
  9. ΒΝΠΔΕ
  10. 2
  11. 53
    • Over 55 million people worldwide are affected by Alzheimer's disease (AD) related and
      other types of dementia; a number expected to double over the next few decades. A slowly
      progressing disease, AD damages irreparably brain multiple regions. Lifestyle choices,
      age, gender and genetics are risk factors, however its cause is still unknown, and only by
      early diagnosis and appropriate therapy can the disease be slowed down. Diagnosis of AD
      includes clinical examination, behavioral and psychiatric evaluation and the detection of
      specific biomarkers related to AD pathology: amyloid plaques, neurofibrillary tangles and
      neuronal injury. The A/T/N (Amyloid/Tau/Neurodegeneration) is used to map these
      biomarkers to the classification of the pathology. Neuroimaging techniques can detect the
      pathopsysiologic and topographic changes in the brain, and especially magnetic resonance
      imaging (MRI) is a commonly used, non invasive, easily accessible examination that
      detects changes in the volume, size, shape and texture of the different brain areas and
      depicts the progression and type of dementia. Machine learning (ML) is applied widely in
      medical application and the neuroimaging field for classification, outlier detection,
      regression and clustering problems, using labeled data in supervised models and unlabeled
      data in unsupervised ones. ML models used on MRI data can detect healthy versus AD
      subjects and predict the course of the disease by detecting changes in the brain. Datasets
      need to be examined, preprocessed and transformed before being fed to an ML model, and
      those choices often affect the performance and predictions. Data is split into two or three
      datasets and fed into the ML models chosen by the researches, then the results are
      evaluated using popular metrics, compared and visualized.


  12. Hellenic Open University
  13. Αναφορά Δημιουργού - Μη Εμπορική Χρήση - Παρόμοια Διανομή 4.0 Διεθνές