Automatic Volumetric Quality Assessment of Diffusion MR Images via Convolutional Neural Network Classifiers

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Artificial Intelligence

Background:

• Diffusion Tensor Imaging (DTI) is crucial for studying brain development and identifying biomarkers.
• Manual Quality Assessment (QA) of large DTI datasets is time-consuming and labor-intensive.
• Automated QA is needed to efficiently process extensive neuroimaging data.

Purpose of the Study:

• To develop a deep learning-based tool for automated, rapid Quality Assessment (QA) of 3D raw diffusion MRI data.
• To create a two-step framework for binary classification ('good' vs 'poor') of DTI volumes.
• To reduce the manual workload associated with DTI data quality control.

Main Methods:

• A 2-step deep learning framework was implemented for automated DTI quality classification.
• Step 1: Two 3D Convolutional Neural Networks (CNNs) with varying input sizes predicted quality labels for sampled Regions of Interest (ROIs).
• Step 2: Novel voting systems aggregated ROI labels to determine the overall DTI volume quality.

Main Results:

• The automated tool achieved high accuracy in classifying DTI data quality.
• One voting system achieved 100% accuracy, while another achieved 98% accuracy on a test set.
• The model was trained and validated on a balanced dataset of 6,940 manually-labeled DTI volumes from 85 subjects.

Conclusions:

• The developed deep learning tool offers a valid and practical solution for automatic DTI quality assessment.
• The automated QA system significantly enhances efficiency in processing large DTI datasets.
• This approach facilitates large-scale neuroimaging studies by streamlining data quality control.