Deep Learning-Enabled Identification of Autoimmune Encephalitis on 3D Multi-Sequence MRI.

Area of Science:

• Artificial Intelligence in Medicine
• Neuroimaging Analysis
• Machine Learning for Diagnostics

Background:

• Autoimmune encephalitis (AE) presents as a noninfectious emergency with severe clinical manifestations.
• Early diagnosis of acute AE is challenging due to limited antibody detection resources.

Purpose of the Study:

• To develop a deep learning (DL) algorithm utilizing multi-sequence magnetic resonance imaging (MRI) for acute AE identification.
• To enhance diagnostic capabilities for AE through advanced computational methods.

Main Methods:

• A retrospective study involving 160 AE patients, 177 herpes simplex virus encephalitis (HSVE) patients, and 184 healthy controls (HC).
• Five DL models were trained on individual or combined MRI sequences (T1 WI, T2 WI, FLAIR, DWI) for classification.
• External validation was performed on 52 patients from a separate site; radiologist performance was also assessed for comparison.

Main Results:

• The fusion DL model achieved superior diagnostic performance across internal and external validation sets, with AUCs ranging from 0.828 to 0.899.
• The fusion model demonstrated significantly higher accuracy (83%) in identifying AE compared to average radiologists (72%).
• The DL algorithm provided robust classification for AE, HSVE, and HC.

Conclusions:

• The developed DL algorithm, leveraging multi-sequence MRI, offers a promising approach for the identification and classification of acute AE.
• This AI-driven method shows potential to improve diagnostic accuracy and efficiency in clinical settings.