Optimising a Simple Fully Convolutional Network for Accurate Brain Age Prediction in the PAC 2019 Challenge.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Biomarker Discovery

Background:

• Brain age prediction from MRI scans is crucial for understanding brain aging.
• Brain-age delta serves as a significant biomarker for brain health.
• The Predictive Analysis Challenge 2019 focused on age prediction from multicentre T1-weighted brain MRIs.

Purpose of the Study:

• To develop and report brain age prediction models for the Predictive Analysis Challenge 2019.
• To evaluate the performance of a Simple Fully Convolutional Neural Network (SFCN) for brain age prediction.
• To assess the utility of various techniques like data augmentation, transfer learning, and bias correction.

Main Methods:

• Utilized a lightweight deep convolutional neural network architecture, Simple Fully Convolutional Neural Network (SFCN).
• Implemented data augmentation, transfer learning, model ensemble, and bias correction techniques.
• Applied the model to multicentre T1-weighted brain MRI datasets.

Main Results:

• Achieved first place in the Predictive Analysis Challenge 2019 for brain age prediction.
• Obtained a Mean Absolute Error (MAE) of 2.90 years without bias removal.
• Achieved an MAE of 2.95 years with bias removal, outperforming competitors significantly.

Conclusions:

• The developed SFCN model demonstrates high accuracy and effectiveness in brain age prediction.
• The combination of deep learning and specific techniques provides a robust approach for neuroimaging analysis.
• The model's performance establishes a new benchmark for predictive analysis in brain aging research.