Classification of Alzheimer’s Disease with 3D CNN-LSTM on MRI Based on Time Difference

Dündar M. S., Yılmaz B.

European Biotechnology Congress, Ljubljana, Slovenya, 4 - 06 Ekim 2023, cilt.7, ss.57-58, (Özet Bildiri)

  • Yayın Türü: Bildiri / Özet Bildiri
  • Cilt numarası: 7
  • Doi Numarası: 10.2478/ebtj-2023-0019
  • Basıldığı Şehir: Ljubljana
  • Basıldığı Ülke: Slovenya
  • Sayfa Sayıları: ss.57-58
  • Erciyes Üniversitesi Adresli: Evet

Özet

In this study, the goal was to design and implement a deep learning model for the analysis of 3D MRI brain scan sequences to differentiate between 22 Alzheimer's Disease (AD) and 18 healthy controls (CN).

The MRI data, which is stored in NIfTI format, was sourced from an organized dataset, cataloging metadata such as image identifiers, patient names, class labels, and sex information. The data extraction procedure was meticulously structured: MRI scans were first segregated by unique subjects, then resized to a consistent target size, and finally organized into sequences for each patient. To effectively capture the longitudinal variations in the MRI scans across 3-4 different time points from each patients, a Long Short-Term Memory (LSTM) network was integrated into a pre-trained 3D ResNet architecture.

Emphasizing the significance of time-differentiated patterns, this combined network was engineered to handle the sequences of 3D images, thereby enhancing its capability to discern between AD and CN based on the temporal dynamics inherent in the MRI sequences. We have achieved accuracy of %94 on classification of AD and CN with %10 higher than classification with 3D ResNet.