Classifying the Acquisition Sequence for Brain MRIs Using Neural Networks on Single Slices.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Deep learning models for MRI analysis are often trained on specific acquisition parameters.
• Real-world MRI data variability hinders the clinical translation of AI research.
• Standardized acquisition sequence data is lacking in clinical practice.

Purpose of the Study:

• To evaluate the feasibility of using convolutional neural networks to classify MRI acquisition sequences from single slices.
• To assess the performance of different neural network complexities for this task.

Main Methods:

• Transfer learning was employed to train three convolutional neural networks on 113 MRI slices.
• Internal validation used 27 slices, followed by external validation on 600 slices across four sequence types.
• Performance was evaluated using categorical accuracy and confusion matrices.

Main Results:

• The neural networks achieved external validation accuracies of 0.79, 0.81, and 0.84.
• Grad-CAM analysis indicated focus on cerebrospinal fluid for T2-weighted slices.
• No distinct focus patterns were observed for other sequence types.

Conclusions:

• Automated classification of MRI acquisition sequences via neural networks is achievable.
• This technology can streamline the automatic labeling of diverse MRI datasets.
• Potential to bridge the gap between AI research and clinical MRI application.