Multi-instance learning attention model for amyloid quantification of brain sub regions in longitudinal cognitive...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Amyloid PET scans are crucial for identifying beta-amyloid deposition in the brain, a hallmark of Alzheimer's disease.
• Current methods may require multiple imaging modalities, increasing complexity and cost.
• Automating the analysis of amyloid PET scans can improve efficiency and accessibility.

Purpose of the Study:

• To develop and validate a deep learning model for automated detection of amyloid deposition using only PET scans.
• To assess the model's performance without the need for co-registered MRI scans.
• To compare the model's accuracy against existing methods using key performance metrics.

Main Methods:

• A deep learning model utilizing multi-instance learning and attention was developed.
• The model was trained and validated on 2647 18F-Florbetapir PET scans from the Alzheimer's Disease Neuroimaging Initiative (ADNI).
• Performance was evaluated using Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) on internal and external datasets (A4 study).

Main Results:

• The model achieved an MAE of 0.0243 and RMSE of 0.0320 on the ADNI test set.
• On the external A4 study dataset, the model achieved an MAE of 0.038 and RMSE of 0.0495.
• The proposed model demonstrated lower MAE and RMSE compared to existing models.

Conclusions:

• The developed deep learning model effectively automates amyloid deposition detection from PET scans without MRI.
• The model shows high accuracy and improved performance, offering a valuable tool for Alzheimer's disease research and diagnosis.
• A user-friendly graphical interface was created for practical application of the model.