Related Experiment Video
Updated: Jun 23, 2025
09:41
Imaging CD19+ B Cells in an Experimental Autoimmune Encephalomyelitis Mouse Model using Positron Emission Tomography
Published on: January 20, 2023
1.8K
Prediction of Seropositivity in Suspected Autoimmune Encephalitis by Use of Radiomics: A Radiological
Jacob Stake1, Christine Spiekers1, Burak Han Akkurt1
1University Clinic for Radiology, University of Münster and University Hospital Münster, Albert-Schweitzer-Campus 1, 48149 Münster, Germany.
Diagnostics (Basel, Switzerland)
|June 19, 2024
Summary
Radiomics and machine learning accurately predict autoantibodies in suspected autoimmune encephalitis (AE) using MRI scans. This approach can expedite diagnosis before lab results are available.
Area of Science:
- Radiology
- Artificial Intelligence
- Neurology
Background:
- Autoimmune encephalitis (AE) diagnosis relies on detecting specific autoantibodies, often requiring time-consuming laboratory tests.
- Early and accurate diagnosis of AE is crucial for timely treatment and improved patient outcomes.
- Predicting seropositivity from neuroimaging could significantly accelerate the diagnostic workflow.
Purpose of the Study:
- To evaluate the efficacy of radiomics and machine learning in predicting autoantibody seropositivity in suspected AE patients.
- To assess the performance of these AI-driven methods using MRI data from the time of symptom onset.
- To explore the potential of radiomics for early AE diagnosis and personalized treatment strategies.
Main Methods:
- Manual segmentation of amygdala regions of interest (ROIs) on T2-weighted MRI scans from 83 AE patients (43 seropositive, 40 seronegative).
- Extraction of 107 radiomic features from segmented ROIs.
- Application of automated machine learning (AutoML) for algorithm selection and recursive feature elimination (RFE) for feature selection.
- Training and evaluation of machine learning models on randomly split training and independent test datasets.
Main Results:
- The radiomics approach achieved a mean Area Under the Curve (AUC) of 0.90, accuracy of 0.83, sensitivity of 0.84, and specificity of 0.82 in predicting seropositivity on independent test data.
- Lasso regression models demonstrated the most promising performance among the evaluated machine learning algorithms.
- The findings indicate a high capability of radiomics and machine learning to predict autoantibody presence from initial MR images.
Conclusions:
- Radiomics-based machine learning is a powerful tool for predicting autoantibody seropositivity in suspected AE cases.
- This AI-driven method can expedite the diagnostic process for AE, even before laboratory test results are finalized.
- Future applications may include AE subtype characterization and personalized medicine approaches based on radiomic profiles.
