Artificial intelligence using neural network architecture for radiology (AINNAR): classification of MR imaging...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Accurate identification of Magnetic Resonance Imaging (MRI) sequences is crucial for clinical diagnosis.
• Confusion in MRI sequence naming can arise post-acquisition, necessitating automated identification methods.

Purpose of the Study:

• To evaluate the capability of deep learning models to automatically classify head MRI sequences.
• To address the challenge of identifying MRI sequences after image data acquisition.

Main Methods:

• Two deep learning classifiers, AlexNet and GoogLeNet, were trained to classify head MRI sequences.
• Data included 78 mild cognitive impairment (MCI) and 78 intracranial hemorrhage (ICH) patients.
• Six MRI sequences were analyzed: T2-weighted, FLAIR, T2*-weighted, DWI, ADC mapping, and TOF-MRA source images.

Main Results:

• GoogLeNet achieved high classification accuracies (93.1%-100%) across different slice types and patient groups.
• AlexNet showed significantly lower classification abilities (60.7%-73.6%) compared to GoogLeNet.
• GoogLeNet demonstrated robust performance irrespective of image morphology or contrast.

Conclusions:

• Deep learning, specifically GoogLeNet, can effectively classify head MRI sequences with minimal training data.
• Automated MRI sequence classification aids in resolving naming confusions and streamlines image data management.