Automatic Phase and Sequence Identification in Gd-EOB-DTPA-Enhanced Liver MRI Using Deep Convolutional and Sequential...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Accurate identification of acquisition sequences in liver MRI is crucial for data curation.
• Gadolinium ethoxybenzyl diethylenetriamine pentaacetic acid (Gd-EOB-DTPA)-enhanced MRI is widely used for liver imaging.
• Manual data organization is time-consuming and prone to errors.

Purpose of the Study:

• To develop and validate a deep learning model for automatic identification of Gd-EOB-DTPA-enhanced liver MRI sequences.
• To enable automated examination-level data curation for multicenter studies.
• To improve the efficiency and accuracy of liver MRI data organization.

Main Methods:

• A deep learning pipeline using ConvNeXt for feature extraction and gated recurrent unit (GRU)-based or transformer-based sequential models was developed.
• Models were trained on series-level 3D image volumes, excluding textual metadata.
• Performance was evaluated on internal and external test sets using examination-level and category-level accuracy metrics.

Main Results:

• The ConvNeXt + GRU model achieved the highest examination-level accuracy.
• Dynamic contrast-enhanced phases were identified with high accuracy across datasets.
• Performance on auxiliary sequences varied, with lower accuracy on external datasets, particularly for T2-weighted imaging.

Conclusions:

• The proposed framework accurately identifies dynamic phases and auxiliary sequences in Gd-EOB-DTPA-enhanced liver MRI.
• Automated identification supports robust examination-level data organization in multicenter settings.
• Further refinement may be needed for auxiliary sequences with high inter-institutional variability.