Disease detection and classification in temporal lobe epilepsy: step-wise versus simultaneous AI decision models in a multisite neuroimaging study

Abstract Diagnostic MRI evaluation of temporal lobe epilepsy (TLE) depends on the subjective visual interpretation of MRI images. These interpretations could be enhanced by quantitative artificial intelligence (AI) support tools. Humans often make sequential and conditional decisions during their radiological interpretations, such as whether an abnormality is present and if present, characterizing the abnormality. It is not known whether it is superior to train AI to treat every decision separately in a similar step-wise manner or to train a model holistically on all decisions simultaneously. Here, we analyzed three large epilepsy MRI datasets [n=3,676, 2,320 people with epilepsy and 1,356 healthy controls (HC)] to perform two tasks: 1) establish the presence of a TLE pattern on MRI and 2) determine TLE pattern lateralization. We compared Step-wise models that independently classify TLE versus HC and lateralize patients as left TLE (L-TLE) or right TLE (R-TLE), against a Simultaneous model trained to distinguish all three classes in a single step. To do this, 3D volumetric T1-weighted images were input into an EfficientNetV2 model multiple times to ensure reproducibility of results. Class prediction, model classification confidence, and saliency maps were output for interpretability. Step-wise models outperformed the Simultaneous model on both tasks (both ps<.001), with an average ∼2.8% accuracy increase for discriminating HC from TLE and an average 12.7% accuracy increase for distinguishing L-TLE from R-TLE. For both the Step-wise and Simultaneous models, important features discriminating TLE from HC included the known TLE limbic pattern involving the hippocampus, para-hippocampal cortical regions, cingulate cortex, and lateral temporal regions. However, there was less concordance between the Step-wise and Simultaneous models for the L-TLE versus R-TLE task (all Fisher’s Zs>10.5, ps<.001); the Step-wise model focused less on subcortical regions such as the thalamus and hippocampus and focused more on distributed cortical pathology. Across the two Step-wise models, 95.1% of TLE patients had accurate classifications in either HC versus TLE and/or L-TLE versus R-TLE tasks. These results included 69.6% of patients being both correctly labeled as TLE and lateralized, 13.9% being correctly labeled TLE but lateralized incorrectly, and 11.6% being lateralized correctly but not detected as TLE. These findings provide evidence that diagnostic tasks with simpler, Step-wise AI models may enhance diagnostic performance and interpretability in clinical workflows. Future AI clinical support tools can leverage this step-wise approach in the early identification of TLE-related structural patterns, supporting timely diagnosis and treatment decisions.

Authors

Institutions

Publication Details

Journal
Brain Communications
Published
2026-06-29
DOI
https://doi.org/10.1093/braincomms/fcag253
Primary Topic
Epilepsy research and treatment
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

Disease detection and classification in temporal lobe epilepsy: step-wise versus simultaneous AI decision models in a multisite neuroimaging study

Patricia Dugan, Orrin Devinsky, Boris C. Bernhardt, Emanuele Bartolini et al.
Brain Communications
Epilepsy research and treatment
article

Disease detection and classification in temporal lobe epilepsy: step-wise versus simultaneous AI decision models in a multisite neuroimaging study

Patricia Dugan, Orrin Devinsky, Boris C. Bernhardt, Emanuele Bartolini, Matteo Lenge, K Schubert, Anees Abrol, R. Guerrini, Erik Kaestner, R Kuzniecky, Terence John O'Brien, Jonathan K. Kleen, Sophia I. Thomopoulos, Ev‐Christin Heide, Meng Law, Nishant Sinha, Kyle Hasenstab, Stefano Meletti, Pascal Martin, Reihaneh Hassanzadeh, Daniel L. Drane, Anna Elisabetta Vaudano, Lucas Scárdua Silva, Fernando Cendes, Antonio Gambardella, Alexandra Parashos, Taha Gholipour, Chantal Depondt, Tobias Bauer, Benjamin Sinclair, Julie Absil, Gian Marco Duma, Sara Larivière, Helena Martins Custodio, Leigh N. Sepeta, Lucy E. Vivash, Luis Concha, André Altmann, Carrie R. McDonald, Jacob Bunyamin, Kathryn A. Davis, Alice Ballerini, Marian Galovic, Donatello Arienzo, Sanjay M. Sisodiya, Richard J. Staba, Angelo Labate, Clarissa Lin Yasuda, Anto Bagić, Paolo Bonanni, Pasquale Striano, Theodor Rüber, Raphaël Christin, Elisa Barbi, Niels K N Focke, Carolina Ferreira-Atuesta, Patrick Moloney, Heath Pardoe, Travis Stoub, Merran R Courtney, Victoria Ives-Deliperi, Marina K M Alvim, Zhiqiang Zhang, Leonardo Bonilha, Ezequiel Gleichgerrcht, Paul M Thompson, Vince D Calhoun, Luca Saba, Jay Sawant
article en

Abstract

Abstract Diagnostic MRI evaluation of temporal lobe epilepsy (TLE) depends on the subjective visual interpretation of MRI images. These interpretations could be enhanced by quantitative artificial intelligence (AI) support tools. Humans often make sequential and conditional decisions during their radiological interpretations, such as whether an abnormality is present and if present, characterizing the abnormality. It is not known whether it is superior to train AI to treat every decision separately in a similar step-wise manner or to train a model holistically on all decisions simultaneously. Here, we analyzed three large epilepsy MRI datasets [n=3,676, 2,320 people with epilepsy and 1,356 healthy controls (HC)] to perform two tasks: 1) establish the presence of a TLE pattern on MRI and 2) determine TLE pattern lateralization. We compared Step-wise models that independently classify TLE versus HC and lateralize patients as left TLE (L-TLE) or right TLE (R-TLE), against a Simultaneous model trained to distinguish all three classes in a single step. To do this, 3D volumetric T1-weighted images were input into an EfficientNetV2 model multiple times to ensure reproducibility of results. Class prediction, model classification confidence, and saliency maps were output for interpretability. Step-wise models outperformed the Simultaneous model on both tasks (both ps<.001), with an average ∼2.8% accuracy increase for discriminating HC from TLE and an average 12.7% accuracy increase for distinguishing L-TLE from R-TLE. For both the Step-wise and Simultaneous models, important features discriminating TLE from HC included the known TLE limbic pattern involving the hippocampus, para-hippocampal cortical regions, cingulate cortex, and lateral temporal regions. However, there was less concordance between the Step-wise and Simultaneous models for the L-TLE versus R-TLE task (all Fisher’s Zs>10.5, ps<.001); the Step-wise model focused less on subcortical regions such as the thalamus and hippocampus and focused more on distributed cortical pathology. Across the two Step-wise models, 95.1% of TLE patients had accurate classifications in either HC versus TLE and/or L-TLE versus R-TLE tasks. These results included 69.6% of patients being both correctly labeled as TLE and lateralized, 13.9% being correctly labeled TLE but lateralized incorrectly, and 11.6% being lateralized correctly but not detected as TLE. These findings provide evidence that diagnostic tasks with simpler, Step-wise AI models may enhance diagnostic performance and interpretability in clinical workflows. Future AI clinical support tools can leverage this step-wise approach in the early identification of TLE-related structural patterns, supporting timely diagnosis and treatment decisions.

Brain Communications
University College Dublin (IE), University of Messina (IT), University of Southern California (US), University of Modena and Reggio Emilia (IT), Rush University Medical Center (US), Montreal Neurological Institute and Hospital (CA), Georgia Institute of Technology (US), The Royal Melbourne Hospital (AU), Université Libre de Bruxelles (BE), Children's National (US), Université de Sherbrooke (CA), University of Bonn (DE), Hofstra University (US), Emory University (US), Medical University of South Carolina (US), University of South Carolina (US), Autonomous University of Queretaro (MX), University of California, Los Angeles (US), University of Pittsburgh (US), University of Cagliari (IT), University of California, San Francisco (US), University of Cologne (DE), Universidade Estadual de Campinas (UNICAMP) (BR), Neurosciences Institute (US), University of Zurich (CH), Florey Institute of Neuroscience and Mental Health (AU), San Diego State University (US), University Hospital Bonn (DE), Istituto Giannina Gaslini (IT), German Center for Neurodegenerative Diseases (DE), Mater Misericordiae University Hospital (IE), Epilepsy Research UK (GB), University of California San Diego (US), Magna Graecia University (IT), NYU Langone Health (US), Monash Alfred Psychiatry Research centre (AU), University Hospital of Zurich (CH), Meyer Children's Hospital (IL), Azienda Ospedaliero-Universitaria di Modena (IT), Meyer Children's Hospital (IT), Universitätsmedizin Göttingen (DE), Fondazione Stella Maris (IT), Center for Translational Research in Neuroimaging and Data Science (US), Second Affiliated Hospital of Nanjing Medical University (CN), Hertie Institute for Clinical Brain Research (DE), Brazilian Institute of Neuroscience and Neurotechnology (BR), IRCCS Eugenio Medea (IT), Epilepsy Foundation (AU), Alfred Health (AU), University of Florence (IT), University College London (GB), Epilepsy Society (GB), Monash University (AU), New York University (US), University of Göttingen (DE), University of Pennsylvania (US), Nanjing Medical University (CN), University of Genoa (IT), Universidad Nacional Autónoma de México (MX)
Reduced inequalities
Openalex Percentile: Top 8%
Epilepsy research and treatment
DOIOpen Access PDFOpenAlex
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.