Machine learning and neuroimaging in neurodegenerative disease diagnosis: a systematic review of voxel-based and surface-based morphometry approaches based on PRISMA framework

Gürlek E., ZAİM GÖKBAY İ.

Neural Computing and Applications, cilt.38, sa.5, 2026 (Scopus) identifier

  • Yayın Türü: Makale / Derleme
  • Cilt numarası: 38 Sayı: 5
  • Basım Tarihi: 2026
  • Doi Numarası: 10.1007/s00521-026-12022-1
  • Dergi Adı: Neural Computing and Applications
  • Derginin Tarandığı İndeksler: Scopus, Compendex, Index Islamicus, INSPEC, zbMATH
  • Anahtar Kelimeler: Deep learning, Machine learning, Neural networks, Neurodegenerative diseases, PRISMA framework, Surface-based morphometry (SBM), Systematic literature review (SLR), Voxel-based morphometry (VBM)
  • İstanbul Üniversitesi Adresli: Evet

Özet

This PRISMA-guided review systematically evaluated the application of machine learning (ML) and deep learning (DL) methodologies to structural Magnetic Resonance Imaging (MRI), specifically utilizing Voxel-Based Morphometry (VBM) and Surface-Based Morphometry (SBM) for ND classification. Our analysis encompassed twelve studies published between 2017 and 2024, focusing on key parameters such as participant demographics, utilized datasets, preprocessing steps, and applied algorithms. To assess the methodological robustness of the included studies, a 20-point Quality Assessment (QA) score was developed, covering data rigor, model development, and validation transparency. The findings indicated significant potential, with algorithms such as Support Vector Machine (SVM) and Extreme Learning Machine (ELM) demonstrating high accuracy, reaching up to 0.96, in the classification of Alzheimer’s and Parkinson’s disease. However, the QA analysis revealed a significant negative correlation () between reporting transparency and performance, suggesting an ’optimistic bias’ in studies with lower methodological rigor. Despite these promising results, the field faced significant challenges, primarily due to small sample sizes—for instance, a limited cohort of for Lewy Body Dementia (LBD)—and a notable lack of consistency in preprocessing techniques, as well as a total absence of public source code sharing (0%), which collectively restricted the generalizability of the models. In conclusion, while the existing research yielded encouraging outcomes, the successful transition to clinical adoption necessitated further innovation in algorithmic development coupled with the incorporation of substantially larger cohorts. To effectively bridge the existing gap between research findings and practical clinical application, the review strongly advocated for the establishment of standardized processing pipelines and a greater commitment to open-data initiatives.