Related Experiment Video
Updated: Aug 18, 2025
06:33
Author Spotlight: Streamlined Brain and Skull Modeling for Enhanced Neurosurgical Planning in NHP Research
Published on: February 9, 2024
1.3K
Adopting transfer learning for neuroimaging: a comparative analysis with a custom 3D convolution neural network
Amira Soliman1, Jose R Chang2,3, Kobra Etminani2
1Center for Applied Intelligent Systems Research (CAISR), Halmstad University, Halmstad, Sweden. amira.soliman@hh.se.
BMC Medical Informatics and Decision Making
|December 8, 2022
Summary
Transfer learning (TL) shows promise for binary classification of neurodegenerative disorders but struggles with multiple diagnoses. A specialized 3D convolutional neural network (CNN) offers better performance and interpretability for complex cases.
Area of Science:
- Neuroimaging
- Deep Learning
- Medical Diagnosis
Background:
- Deep learning (DL) advances neuroimaging for neurodegenerative disorder diagnosis.
- Limited labeled data in medical domains hinders DL application.
- Transfer learning (TL) uses pre-trained networks on natural images to overcome data scarcity.
Purpose of the Study:
- To evaluate specialized and TL models for classifying neurodegenerative disorders.
- To compare model performance using 3D 18F-FDG-PET brain scans.
Main Methods:
- Utilized 3D 18F-FDG-PET brain scan data.
- Compared performance of transfer learning (TL) models with specialized convolutional neural networks (CNNs).
Main Results:
- TL models were suboptimal for classifying more than two neurodegenerative disorders.
- Specialized CNN models demonstrated superior performance in multi-disorder classification.
- Specialized CNNs offered better interpretability of diagnostic predictions.
Conclusions:
- TL excels in binary classification but is less effective for multiple disorder detection.
- Custom 3D CNNs perform comparably to TL for binary classification.
- Custom 3D CNNs outperform TL models in diagnosing multiple neurodegenerative disorders and provide enhanced model explainability.
