A Deep Convolutional Neural Network for Annotation of Magnetic Resonance Imaging Sequence Type

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• The increasing volume of medical imaging data necessitates automated annotation for clinical research.
• Magnetic resonance (MR) image annotation is challenging due to non-standard labeling of sequence types.
• Lack of automated annotation infrastructure hinders the aggregation of large medical image collections.

Purpose of the Study:

• To train a deep neural network for automated annotation of MR image sequence types in brain tumor patients.
• To improve the efficiency and accuracy of medical image data management for research.

Main Methods:

• A deep neural network was trained on 14,400 2D MR brain images.
• The dataset included four common sequence types: T1-weighted (T1W), T1-weighted post-gadolinium contrast (T1Gd), T2-weighted (T2W), and FLAIR.
• Images were acquired across various parameters and manufacturers, independent of patient demographics or diagnosis.

Main Results:

• The deep learning model achieved 99% accuracy in predicting MR image sequence types on the test set.
• The model demonstrated excellent performance in classifying T1W, T1Gd, T2W, and FLAIR sequences.
• Results highlight the effectiveness of deep learning for neuroimaging annotation.

Conclusions:

• Deep learning models are highly effective for annotating brain tumor MR image sequence types.
• Automated annotation using deep learning can significantly support clinical research.
• This technology facilitates efficient management and utilization of large medical image databases.