AI&ML in MRI for the lateralization of temporal lobe epilepsy: a systematic review and meta-analysis
Poster No:
1475
Submission Type:
Abstract Submission
Authors:
Judy Chen1, Natalie Chen2, Yigu Zhou3, Sienna Armstrong4, Lorenzo Caciagli5, Andrea Bernasconi1, Neda Bernasconi1, Boris Bernhardt6
Institutions:
1Montreal Neurological Institute and Hospital, Montreal, QC, 2University of Toronto, Toronto, ON, 3Montreal Neurological Institute, Montreal, Quebec, 4McGill University, Montreal, QC, 5University of Bern, Bern, Bern, 6Montreal Neurological Institute and Hospital, Montreal, Quebec
First Author:
Co-Author(s):
Introduction:
Accurate lateralization of temporal lobe epilepsy (TLE) is critical for drug-resistant patients undergoing surgery but is often difficult to achieve due to conflicting evidence in presurgical work-up.1 Artificial intelligence (AI) and machine learning (ML) methods have been increasingly leveraged to analyze brain magnetic resonance imaging (MRI) for this purpose, but their true accuracy and effectiveness in supporting clinical decision making remains unclear.2 The aim of this meta-analysis was to synthesize current AI/ML models used in MRI and assess their performance in lateralizing TLE.
Methods:
MEDLINE and Embase databases were searched for original research articles and yielded 2606 publications after removing duplicates. Studies were included if they used any AI/ML model trained on (but not limited to) MRI images to classify laterality and reported any evaluation metric (sensitivity/specificity, AUC under ROC, or total accuracy score). Articles were excluded if they were conference abstracts without a full-text publication or had a patient cohort of n<10. Information regarding the publication (year, first author), MRI (Tesla, modality), AI/ML algorithm (model type, input), demographics (patient/control cohort size, sex, epilepsy duration, age of onset, lesion laterality), and algorithm performance metrics (training and test cohort: Sn/Sp, AUC, accuracy score) was extracted from all articles. The risk of bias of each included study was independently assessed by two reviewers using the Prediction model Risk Of Bias Assessment Tool (PROBAST) tool.3
Results:
Abstract and title screening yielded 328 publications, and 51 studies were included. The majority of studies reported an accuracy score (n=29) and were thus included for the meta-analysis; 1440 TLE patients in total (left-sided = 761; mean +/- SD age: 35.4 +/- 5.2, age of onset: 16.3+/-4.6). The accuracy scores in relation to year of study publication, MRI modality, and size of cohort show increasing publications after 2015 with larger cohort sizes and more multimodal MRI used (Fig 2A). The ML algorithms include support vector machine (n=16), any type of discriminant analysis (n=10), logistic regression (n=1), random forest (n=1), and k-means clustering (n=1). Preliminary analyses show that the overall accuracy of AI/ML methods to lateralize patients is 91% (95% CI 87-93%) (Fig 2A). Influence diagnostic plots show general concordance between studies across several influence measures (Fig 2B). Leave-one-out meta-analysis also show highly concordant effect sizes, demonstrating robust accuracies across different modalities and AI/ML algorithms (Fig 2B).
Conclusions:
AI/ML methods are highly accurate in lateralizing temporal lobe epilepsy (TLE). They have an overall accuracy rate of over 90%, with highly consistent and robust results across various study designs and algorithm types. AI/ML on multimodal MRI images can be a powerful aid for the presurgical assessment of drug-resistant TLE patients.
Disorders of the Nervous System:
Neurodevelopmental/ Early Life (eg. ADHD, autism) 2
Modeling and Analysis Methods:
Classification and Predictive Modeling 1
Diffusion MRI Modeling and Analysis
fMRI Connectivity and Network Modeling
Keywords:
ADULTS
Data analysis
Epilepsy
FUNCTIONAL MRI
Machine Learning
Meta- Analysis
MRI
STRUCTURAL MRI
White Matter
WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC
1|2Indicates the priority used for review
·Figure 1. PRISMA reporting. MEDLINE and Embase databases were searched for original research articles that yielded 2606 publications after removing duplicates. Studies were included if they used any A
·Figure 2. Meta-analysis and evaluation of heterogeneity. A | The accuracy scores in relation to year of study publication, MRI modality, and size of cohort show increasing publications after 2015 with
Provide references using author date format
2. Cendes, et al. “Artificial Intelligence Applications in the Imaging of Epilepsy and its Comorbidities: Present and Future” Epilepsy Currents 2022. 22(2):91-96
3. Wolff, Robert F., et al. "PROBAST: a tool to assess the risk of bias and applicability of prediction model studies." Annals of internal medicine 170.1 (2019): 51-58.