About

Richard M. Leahy is the Arthur G. Settle Trust Endowment for USC Leonard Silverman Chair and Professor of Electrical and Computer Engineering, Biomedical Engineering, and Radiology at USC. He holds a Ph.D. in Electrical Engineering from the University of Newcastle-Upon-Tyne, obtained in 1984, and a Bachelor's Degree in Electrical and Electronic Engineering from the same university. Dr. Leahy's research interests focus on the application of signal and image processing theory to the formation and analysis of biomedical images. His group is involved in developing computational methods for positron emission tomography (PET) image formation with applications in clinical oncology and gene expression imaging in small animals. He has also contributed significantly to the development of inverse methods for spatio-temporal imaging of neural activity from magnetoencephalographic (MEG) and electroencephalographic (EEG) data, distributing his work as a freely available software package called BrainStorm. Additionally, his research includes the analysis of anatomical imagery, particularly in automated segmentation and labeling of neuroanatomical images and extracting surface representations of the cerebral cortex from volumetric data. Dr. Leahy has held various academic and professional roles, including Director of the Signal and Image Processing Institute at USC, and has been involved with UCLA's Crump Institute for Molecular Imaging and Laboratory of Neuroimaging. His work is supported by multiple national institutes, and he has received numerous awards and honors for his contributions to medical imaging and biomedical engineering.

Research topics

• Artificial Intelligence
• Psychology
• Computer Science
• Neuroscience
• Biology
• Medicine
• Computer vision
• Cartography
• Anatomy
• Mathematics
• Evolutionary biology
• Geography

Selected publications

• A Point Process Model of Skin Conductance Responses in a Stroop Task for Predicting Depression and Suicidal Ideation

  2026-04-21

  articleOpen access

  Accurate identification of mental health biomarkers can enable earlier detection and objective assessment of compromised mental well-being. In this study, we analyze electrodermal activity recorded during an Emotional Stroop task to capture sympathetic arousal dynamics associated with depression and suicidal ideation. We model the timing of skin conductance responses as a point process whose conditional intensity is modulated by task-based covariates, including stimulus valence, reaction time, and response accuracy. The resulting subject-specific parameter vector serves as input to a machine learning classifier for distinguishing individuals with and without depression. Our results show that the model parameters encode meaningful physiological differences associated with depressive symptomatology and yield superior classification performance compared to conventional feature extraction methods.

  PublisherOA PDFDOI

• Multi-Compartment Volume Conductor with Complete Electrode Model: Simulated Stereo-EEG Source Localization using Brainstorm-Zeffiro Plugin

  arXiv (Cornell University) · 2026-01-31

  articleOpen access

  This study introduces a novel integration of the Brainstorm (BST) software and the Zeffiro Interface (ZI) to enable whole-head, multi-compartment volume conductor modeling for electroencephalography (EEG) source imaging, with a particular focus on stereotactic EEG applications. We present the BST-2-ZI plugin, a MATLAB-based tool that facilitates seamless transfer of tissue segmentations and anatomical atlases from BST into ZI for finite element (FE) mesh generation as well as forward and inverse modeling. The generated FE meshes support variable spatial resolution and implement the complete electrode model (CEM), allowing for precise modeling of both invasive depth electrodes and non-invasive scalp electrodes. Using the ICBM152 template and synthetic source simulation, we demonstrate the end-to-end pipeline from MRI data to lead field (LF) computation and source localization in a stereotactic EEG (stereo-EEG) setting. Our numerical experiments highlight the capability of the pipeline to accurately model multi-compartment head geometry and conductivity with a stereotactic CEM-based electrode configuration. Our preliminary source localization results show how a synthetic stereo-EEG probe corresponding to a bidirectional deep brain stimulation (DBS) probe with four omnidirectional contacts can, in principle, be coupled with scalp electrodes to improve source localization in its vicinity.

  PublisherOA PDF

• Time-resolved EEG decoding reveals altered neural dynamics of affective semantic evaluation in depression and suicidality

  Communications Biology · 2026-04-30

  articleOpen accessSenior author

  Depression and suicidality are associated with systematic alterations in cognitive and emotional processing, yet the spatiotemporal neural dynamics underlying these changes during affective task engagement remain poorly characterized. We investigate the time-resolved neural representations of affective semantic processing using multivariate decoding of 64-channel electroencephalography (EEG), while participants (N = 137) perform Sentence Evaluation task using emotionally salient, self-referential statements. Reliable condition-dependent neural discriminability emerges with peak decoding accuracy between 300-600 ms, a time window corresponding to affective semantic evaluation, contextual updating processes, and conflict monitoring. Relative to healthy controls, individuals with depression and suicidal ideation show earlier onset, longer duration, and greater amplitude decoding responses, along with broader cross-temporal generalization and enhanced contributions from frontocentral and parietotemporal components. These results indicate systematic group-level differences in the spatiotemporal dynamics and stability of neural representations during emotionally salient semantic evaluation, contributing to a characterization of neurocognitive processes associated with depression and suicidality.

  PublisherDOI

• Chronic Anemia Patients Demonstrate Diffuse Demyelination

  American Journal of Hematology · 2026-05-14

  article

  Chronic anemia is observed in individuals with sickle cell disease (SCD) and thalassemia syndromes. It has been associated with a range of neurological complications, particularly progressive silent cerebral infarcts in brain regions with high oxygen extraction-specifically in vascular watershed areas-suggesting that regional tissue hypoxia may play a causal role. However, recent work in sickle cell mice reveals widespread white matter demyelination and chronic neuroinflammation superimposed upon regional ischemia. In light of these findings, we utilized high-fidelity diffusion imaging and modeling techniques to identify predictors of white matter damage in human subjects diagnosed with SCD (n = 76) and thalassemia (n = 20) compared to healthy individuals (n = 32). Our results demonstrate that white matter damage extended beyond vascular watershed areas in chronically anemic subjects and had MRI changes characteristic of demyelination. These findings were proportional to hemoglobin levels and largely disappeared after controlling for anemia severity. However, patients with SCD exhibited small but significant residual white matter derangements not seen in those with thalassemia. These residual abnormalities disappeared after LDH or reticulocyte count were included as markers of hemolytic rate. From a functional perspective, neuropsychological processing speed was correlated with white matter integrity in chronic anemia subjects, with stronger associations seen in patients with SCD. Taken together, these results demonstrate that chronic anemia is associated with widespread white matter demyelination that cannot be explained by regional blood flow variation and is proportional to anemia severity. Patients with SCD may have more severe disease and functional consequences than patients with thalassemia.

  PublisherDOI

• Neural Responses to Affective Sentences Reveal Signatures of Depression

  Translational Psychiatry · 2026-05-15

  preprintOpen access

  Major Depressive Disorder (MDD) is a highly prevalent mental health condition, and a deeper understanding of its neurocognitive foundations is essential for identifying how core functions such as emotional and self-referential processing are affected. We investigate how depression symptoms alters the temporal dynamics of emotional processing by measuring neural responses to self-referential affective sentences using surface electroencephalography (EEG) in healthy and depressed individuals. Our results reveal significant group-level differences in neural activity during sentence viewing, suggesting disrupted integration of emotional and self-referential information in depression. Deep learning model trained on these responses achieves an area under the receiver operating curve (AUC) of 0.72 in distinguishing healthy from depressed participants, and 0.65 in differentiating depressed subgroups with and without suicidal ideation symptoms. Spatial ablations highlight anterior electrodes associated with semantic and affective processing as key contributors. These findings suggest stable, stimulus-driven neural signatures of depression symptoms that may inform future screening tools.

  PublisherOA PDFDOI

• Predicting Post-Traumatic Epilepsy from Clinical Records using Large Language Model Embeddings

  arXiv (Cornell University) · 2026-04-16

  preprintOpen accessSenior author

  Objective: Post-traumatic epilepsy (PTE) is a debilitating neurological disorder that develops after traumatic brain injury (TBI). Early prediction of PTE remains challenging due to heterogeneous clinical data, limited positive cases, and reliance on resource-intensive neuroimaging data. We investigate whether routinely collected acute clinical records alone can support early PTE prediction using language model-based approaches. Methods: Using a curated subset of the TRACK-TBI cohort, we developed an automated PTE prediction framework that implements pretrained large language models (LLMs) as fixed feature extractors to encode clinical records. Tabular features, LLM-generated embeddings, and hybrid feature representations were evaluated using gradient-boosted tree classifiers under stratified cross-validation. Results: LLM embeddings achieved performance improvements by capturing contextual clinical information compared to using tabular features alone. The best performance was achieved by a modality-aware feature fusion strategy combining tabular features and LLM embeddings, achieving an AUC-ROC of 0.892 and AUPRC of 0.798. Acute post-traumatic seizures, injury severity, neurosurgical intervention, and ICU stay are key contributors to the predictive performance. Significance: These findings demonstrate that routine acute clinical records contain information suitable for early PTE risk prediction using LLM embeddings in conjunction with gradient-boosted tree classifiers. This approach represents a promising complement to imaging-based prediction.

  PublisherDOI

• Optimizing electrode placement and information capacity for local field potentials in cortex

  NeuroImage · 2026-01-21

  articleOpen access

  Recent neurosurgery advancements include improved stereotactic targeting and increased density and specificity of electrophysiological evaluation. This study introduces a subject-specific, in silico modeling tool for optimizing electrode placement and maximizing coverage with a variety of devices. The basis for optimization is the Shannon-Hartley information capacity of field potentials derived from dipolar sources. The approach integrates subject-specific MRI data with finite element modeling (FEM) used to simulate the sensitivity of subdural and intracortical devices. Sensitivity maps, or lead fields, from these models enable the comparison of different electrode placements, contact sizes, contact configurations, and substrate properties, which are often overlooked factors. One key tool is a genetic algorithm that optimizes electrode placement by maximizing information capacity. Another is a sparse sensor method, Sparse Electrode Placement for Input Optimization (SEPIO), that selects the best sensor subsets for accurate source classification. We demonstrate several use cases for clinicians, engineers, and researchers. Overall, these open-source tools provide a quantitative framework to select devices from a neurosurgical armament and to optimize device and contact placement. Using these tools may help refine electrode coverage with low channel count devices while minimizing the burden of invasive surgery. The study demonstrates that optimized electrode placement significantly improves the information capacity and signal quality of local field potential (LFP) recordings. The tools developed offer a valuable approach for refining neurosurgical techniques and enhancing the design of neural implants.

  PublisherDOI

• Predicting Post-Traumatic Epilepsy from Clinical Records using Large Language Model Embeddings

  ArXiv.org · 2026-04-16

  articleOpen accessSenior author

  Objective: Post-traumatic epilepsy (PTE) is a debilitating neurological disorder that develops after traumatic brain injury (TBI). Early prediction of PTE remains challenging due to heterogeneous clinical data, limited positive cases, and reliance on resource-intensive neuroimaging data. We investigate whether routinely collected acute clinical records alone can support early PTE prediction using language model-based approaches. Methods: Using a curated subset of the TRACK-TBI cohort, we developed an automated PTE prediction framework that implements pretrained large language models (LLMs) as fixed feature extractors to encode clinical records. Tabular features, LLM-generated embeddings, and hybrid feature representations were evaluated using gradient-boosted tree classifiers under stratified cross-validation. Results: LLM embeddings achieved performance improvements by capturing contextual clinical information compared to using tabular features alone. The best performance was achieved by a modality-aware feature fusion strategy combining tabular features and LLM embeddings, achieving an AUC-ROC of 0.892 and AUPRC of 0.798. Acute post-traumatic seizures, injury severity, neurosurgical intervention, and ICU stay are key contributors to the predictive performance. Significance: These findings demonstrate that routine acute clinical records contain information suitable for early PTE risk prediction using LLM embeddings in conjunction with gradient-boosted tree classifiers. This approach represents a promising complement to imaging-based prediction.

  PublisherOA PDF

• Multi-Compartment Volume Conductor with Complete Electrode Model: Simulated Stereo-EEG Source Localization using Brainstorm-Zeffiro Plugin

  Open MIND · 2026-01-31

  preprint

  This study introduces a novel integration of the Brainstorm (BST) software and the Zeffiro Interface (ZI) to enable whole-head, multi-compartment volume conductor modeling for electroencephalography (EEG) source imaging, with a particular focus on stereotactic EEG applications. We present the BST-2-ZI plugin, a MATLAB-based tool that facilitates seamless transfer of tissue segmentations and anatomical atlases from BST into ZI for finite element (FE) mesh generation as well as forward and inverse modeling. The generated FE meshes support variable spatial resolution and implement the complete electrode model (CEM), allowing for precise modeling of both invasive depth electrodes and non-invasive scalp electrodes. Using the ICBM152 template and synthetic source simulation, we demonstrate the end-to-end pipeline from MRI data to lead field (LF) computation and source localization in a stereotactic EEG (stereo-EEG) setting. Our numerical experiments highlight the capability of the pipeline to accurately model multi-compartment head geometry and conductivity with a stereotactic CEM-based electrode configuration. Our preliminary source localization results show how a synthetic stereo-EEG probe corresponding to a bidirectional deep brain stimulation (DBS) probe with four omnidirectional contacts can, in principle, be coupled with scalp electrodes to improve source localization in its vicinity.

  DOI

• High-Resolution Directional Depth Electrodes: Open-Source FEM Lead-Field Modeling, Characterization, and Validation

  ArXiv.org · 2025-08-16

  preprintOpen access

  Depth electrodes used in stereoelectroencephalography (sEEG) and deep-brain stimulation (DBS) are essential tools for neural recording and stimulation. Traditional designs have limited spatial resolution, typically 8 to 16 cylindrical contacts (0.8 to 1.0 mm diameter) along a 5 to 10 cm shaft, restricting recordings from small or localized populations. Recent high-density, directional electrodes (HDsEEG) enable finer localization of local field potentials (LFPs) and spike timing. Yet, characterizing their directional sensitivity and validating modeling tools for lead field (LF) analysis remain critical. We compare finite element method (FEM) LF modeling of a novel HD-sEEG electrode using two tools: a commercial solver (ANSYS) and an open-source pipeline (Brainstorm-DUNEuro). Goals: (i) validate against analytical solutions, (ii) assess solver differences, and (iii) characterize HDsEEG directional sensitivity. LFs were modeled in simple and bio-relevant scenarios. Using Helmholtz reciprocity, we computed LFs by (a) electrode-based stimulation with ANSYS and (b) source-based recording with Brainstorm-DUNEuro. First, a multi-sphere head model with known solutions tested the solver's accuracy. Next, an HDsEEG electrode in a homogeneous conductor was simulated. Directional effects were assessed by comparing sensitivity with vs. without the insulating substrate. Source localization performance was also compared between HDsEEG and standard electrodes. Both solvers closely matched analytic solutions. In realistic settings, LF distributions were highly similar. Modeling showed clear directional sensitivity: contacts facing a source had higher sensitivity than those shadowed, reflecting a substrate shielding effect that vanished when the substrate was conductive. Crucially, HDsEEG improved source localization, as voltage differences across contacts provided robust directional LF information.

  PublisherOA PDF

Recent grants

• NIH Grant R13EB003414

  NIH · $20k · 2005

• BrainSuite: Software for Analysis and Visualization of Multimodal Brain Imaging Data

  NIH · $4.6M · 2011–2023

• NIH Grant R01CA059794

  NIH · $1.9M · 2002

• A Brain Atlas for Mapping Connectivity in Focal Epilepsy

  NIH · $2.7M · 2015–2021

• NIH Grant R01EB000363

  NIH · $1.4M · 2006

Frequent coauthors

• Anand A. Joshi

  156 shared

• John C. Mosher

  103 shared

• David W. Shattuck

  Brain Mapping Foundation

  83 shared

• Dimitrios Pantazis

  Massachusetts Institute of Technology

  70 shared

• Sylvain Baillet

  McGill University

  58 shared

• Jian Li

  Massachusetts General Hospital

  57 shared

• Simon R. Cherry

  University of California, Davis

  55 shared

• Quanzheng Li

  Harvard University

  48 shared

Education

• Ph.D., Electrical Engineering

  University of Southern California

  1990

• M.S., Electrical Engineering

  University of Southern California

  1986

• B.S., Electrical Engineering

  University of Southern California

  1984

Awards & honors

• 2011 International Society for Functional Source Imaging (IS…
• 2010 IEEE Nuclear and Plasma Sciences Society Edward J Hoffm…
• 2009 Institute of Physics / Physics Med. Biol Roberts Prize…
• 2009 IEEE Nucl. Sci. Sypm./Med. Imag. Conf Best Student Post…
• 2008 Institute of Physics Paper award