Pegah Ahadian
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Pegah Ahadian has a total of 27 co-authors. Here are the top co-authors by number of shared publications:
- Qiang Guan
Kent State University
11 shared publications - Ali Mehrban
School of Electrical and Electronic Engineering Newcastle University Newcastle UK
2 shared publications - Angela Fragano
Division of Surgical Oncology, Department of Otolaryngology-Head and Neck Surgery, Mike Toth Head and Neck Cancer Research Center, Mass Eye and Ear, Boston, Massachusetts
2 shared publications - Karl Wesley Kosko
Kent State University
2 shared publications - Mingrui Yang
Program of Advanced Musculoskeletal Imaging (PAMI), Cleveland Clinic Cleveland Ohio USA
2 shared publications - Sophia Z. Shalhout
Department of Otolaryngology-Head and Neck Surgery, Division of Surgical Oncology, Mike Toth Cancer Research Center, Massachusetts Eye and Ear, Mass General Brigham, Boston, USA
2 shared publications
Latest Publications
- Open Access
- 24 September 2026
- Published by Springer Nature in Scientific Reports
Development and validation of machine learning-based prediction of depression progression using EHR data: a multi-institutional retrospective cohort study
Depression is common and often difficult to monitor for clinical worsening in real-world settings where structured symptom scales are inconsistently captured. Electronic health records allow large cohort analyses. Diagnostic-code based definitions offer a practical way to identify patients at risk for progression. We used a multi-institutional dataset (MedStar Health) to study adults with mild depression and evaluated whether machine learning models could predict progression to moderate or severe depression within two years. The cohort included 803 individuals with two years of continuous follow-up. A broad set of demographics, clinical, laboratory, socioeconomic, and healthcare utilization features was available. Models included logistic regression, random forest, gradient boosted trees, and a deep neural network (DNN). Gradient boosted trees showed balanced performance, defined as a trade-off between sensitivity and specificity with a modest improvement over baseline prevalence, while logistic regression demonstrated the strongest overall discrimination and generalization. The DNN showed lower accuracy than null classifier. These findings demonstrate that real-world EHR data can support risk prediction of depression worsening with stable and clinically relevant performance. These models may support earlier identification of patients who warrant closer monitoring or targeted follow-up.
- 24 September 2026
- Published by Springer Nature in Scientific Reports
Open AccessDevelopment and validation of machine learning-based prediction of depression progression using EHR data: a multi-institutional retrospective cohort study
Depression is common and often difficult to monitor for clinical worsening in real-world settings where structured symptom scales are inconsistently captured. Electronic health records allow large cohort analyses. Diagnostic-code based definitions offer a practical way to identify patients at risk for progression. We used a multi-institutional dataset (MedStar Health) to study adults with mild depression and evaluated whether machine learning models could predict progression to moderate or severe depression within two years. The cohort included 803 individuals with two years of continuous follow-up. A broad set of demographics, clinical, laboratory, socioeconomic, and healthcare utilization features was available. Models included logistic regression, random forest, gradient boosted trees, and a deep neural network (DNN). Gradient boosted trees showed balanced performance, defined as a trade-off between sensitivity and specificity with a modest improvement over baseline prevalence, while logistic regression demonstrated the strongest overall discrimination and generalization. The DNN showed lower accuracy than null classifier. These findings demonstrate that real-world EHR data can support risk prediction of depression worsening with stable and clinically relevant performance. These models may support earlier identification of patients who warrant closer monitoring or targeted follow-up.
- 11 July 2026
- Published by Springer Nature
- p. 71-90
Long Short-Term Memory Networks for Solving Delay Differential Equations with Non-local Derivatives
- 11 July 2026
- Published by Springer Nature
- p. 71-90
Long Short-Term Memory Networks for Solving Delay Differential Equations with Non-local Derivatives
- Open Access
- 6 April 2026
- Published by MDPI AG in Preprints.org
A Discordance-Aware Multimodal Framework with Multi-Agent Clinical Reasoning
Background: Knee osteoarthritis (OA) frequently exhibits discordance between structural damage observed in imaging and patient-reported symptoms such as pain. This mismatch complicates clinical interpretation and patient stratification and remains insufficiently modeled in existing decision-support systems. Methods: We propose a discordance-aware multimodal framework that combines machine learning prediction models with a tool-grounded multi-agent reasoning system. Using baseline data from the FNIH Osteoarthritis Biomarkers Consortium (600 knees), we trained multimodal models to predict two progression tasks: (i) joint-space-loss-only progression versus non-progression and (ii) pain-only progression versus non-progression. The predictive system integrates three modality- specific experts: a CatBoost tabular model using demographic, radiographic, MRI-derived scalar, and biomarker features; MRI image embeddings extracted using a ResNet18 backbone; and X-ray embeddings derived from the same architecture. Expert predictions are fused using a stacking ensemble. Residual-based models estimate expected pain from structural features, enabling the computation of a pain–structure discordance score between observed and expected symptoms. A multi-agent reasoning layer interprets these signals to assign clinically interpretable OA phenotypes and generate phenotype-specific management recommendations. Results: Using 5-fold stratified cross-validation with out-of-fold evaluation, the full multimodal stacking model combining tabular variables, MRI and X-ray embeddings, and biochemical biomarkers achieved the best performance. For the JSL-only progression task, the model achieved AUC 0.702. For pain-only progression, the model achieved AUC 0.611. Imaging embeddings alone provided limited predictive signal, whereas clinically interpretable radiographic and MRI scalar features contributed stronger discrimination. Multimodal fusion improved performance by integrating complementary structural and biochemical information. Conclusions: Multimodal fusion of tabular clinical variables, imaging-derived features, deep image embeddings, and biochemical biomarkers improves structural progression prediction in knee OA. By coupling this prediction layer with explicit pain–structure discordance modeling and a tool-grounded multi-agent reasoning framework, the proposed architecture supports interpretable phenotype assignment and structured clinical decision support for osteoarthritis management.
- 6 April 2026
- Published by MDPI AG in Preprints.org
Open AccessA Discordance-Aware Multimodal Framework with Multi-Agent Clinical Reasoning
Background: Knee osteoarthritis (OA) frequently exhibits discordance between structural damage observed in imaging and patient-reported symptoms such as pain. This mismatch complicates clinical interpretation and patient stratification and remains insufficiently modeled in existing decision-support systems. Methods: We propose a discordance-aware multimodal framework that combines machine learning prediction models with a tool-grounded multi-agent reasoning system. Using baseline data from the FNIH Osteoarthritis Biomarkers Consortium (600 knees), we trained multimodal models to predict two progression tasks: (i) joint-space-loss-only progression versus non-progression and (ii) pain-only progression versus non-progression. The predictive system integrates three modality- specific experts: a CatBoost tabular model using demographic, radiographic, MRI-derived scalar, and biomarker features; MRI image embeddings extracted using a ResNet18 backbone; and X-ray embeddings derived from the same architecture. Expert predictions are fused using a stacking ensemble. Residual-based models estimate expected pain from structural features, enabling the computation of a pain–structure discordance score between observed and expected symptoms. A multi-agent reasoning layer interprets these signals to assign clinically interpretable OA phenotypes and generate phenotype-specific management recommendations. Results: Using 5-fold stratified cross-validation with out-of-fold evaluation, the full multimodal stacking model combining tabular variables, MRI and X-ray embeddings, and biochemical biomarkers achieved the best performance. For the JSL-only progression task, the model achieved AUC 0.702. For pain-only progression, the model achieved AUC 0.611. Imaging embeddings alone provided limited predictive signal, whereas clinically interpretable radiographic and MRI scalar features contributed stronger discrimination. Multimodal fusion improved performance by integrating complementary structural and biochemical information. Conclusions: Multimodal fusion of tabular clinical variables, imaging-derived features, deep image embeddings, and biochemical biomarkers improves structural progression prediction in knee OA. By coupling this prediction layer with explicit pain–structure discordance modeling and a tool-grounded multi-agent reasoning framework, the proposed architecture supports interpretable phenotype assignment and structured clinical decision support for osteoarthritis management.
- Open Access
Development and Validation of Machine Learning-Based Prediction of Depression Progression Using EHR Data: A Multi-Institutional Retrospective Cohort Study
Background: Depression is a leading cause of global disability. Timely identification of patients at risk for clinical worsening remains a major challenge. Electronic health records (EHRs) facilitate large-scale, real-world analyses of disease trajectories. However, standardized symptom scale data such as the Patient Health Questionnaire-9 are often unavailable or recorded only as unstructured text. In this context, International Classification of Diseases (ICD10) diagnostic-code based severity progression provides a pragmatic alternative for developing predictive tools to identify worsening depression. Objective: We aim to develop and evaluate machine-learning and deep-learning models for predicting ICD10-defined progression from mild to moderate/severe depression using EHR data curated by the MedStar Health Research Institute (MHRI). Methods: We conducted a multi-institutional retrospective cohort analysis using the MHRI EHR database, which integrates data from 10 hospitals and 300 outpatient sites across the mid-Atlantic. Adults (≥18 years) with an initial ICD10 diagnosis of mild depression between 2017 and 2023 were included (N=2131). Nonprogressors were defined as patients whose mild major depressive disorder remained mild for 24 months (N=270). Progressors were defined as patients who developed moderate or severe ICD10 depression within 24 months of the index diagnosis (N=533). Data were stratified and split into (60%) training, (20%) validation, and (20%) test subsets. A heterogeneous feature set spanning demographics, healthcare utilization, socioeconomic indices, diagnostic context, and laboratory measurements were available. Logistic regression utilized elastic net regularization with fivefold cross validation, and random forest hyperparameters were tuned by grid search. XGBoost, CatBoost, and a deep neural network (DNN) were trained with standard learning rate, depth, class weighting, and early stopping. A deterministic top model selection framework applied prespecified thresholds of sensitivity at least 0.70 and AUC at least 0.70, and composite rankings integrated accuracy, sensitivity, specificity, and the overfitting gap. Results: The analytic cohort included 803 patients with complete two-year follow-up. Under the selection criteria, the DNN failed to meet the AUC threshold (0.671) and was excluded. Among the remaining models, XGBoost achieved the top composite score (accuracy = 0.72; AUC = 0.776; sensitivity = 0.77; specificity = 0.63; overfit gap = 0.112). Logistic regression ranked second (accuracy = 0.71; AUC = 0.797; sensitivity = 0.79; specificity = 0.61; overfit gap = 0.052), followed by CatBoost and random forest, the latter penalized for overfitting (gap = 0.278). The TinyLlama audit note, generated through a local Hugging Face pipeline, confirmed XGBoost as the most balanced model. Conclusions: Using EHR data from a multi-institutional regional health system, we developed and validated machine-learning models that predicted progression of depression. XGBoost demonstrated the most reliable composite performance. These findings support the feasibility of leveraging socioeconomic and EHR data to predict worsening depression and emphasize the importance of transparent model-selection frameworks for trustworthy clinical artificial intelligence.
Open AccessDevelopment and Validation of Machine Learning-Based Prediction of Depression Progression Using EHR Data: A Multi-Institutional Retrospective Cohort Study
Background: Depression is a leading cause of global disability. Timely identification of patients at risk for clinical worsening remains a major challenge. Electronic health records (EHRs) facilitate large-scale, real-world analyses of disease trajectories. However, standardized symptom scale data such as the Patient Health Questionnaire-9 are often unavailable or recorded only as unstructured text. In this context, International Classification of Diseases (ICD10) diagnostic-code based severity progression provides a pragmatic alternative for developing predictive tools to identify worsening depression. Objective: We aim to develop and evaluate machine-learning and deep-learning models for predicting ICD10-defined progression from mild to moderate/severe depression using EHR data curated by the MedStar Health Research Institute (MHRI). Methods: We conducted a multi-institutional retrospective cohort analysis using the MHRI EHR database, which integrates data from 10 hospitals and 300 outpatient sites across the mid-Atlantic. Adults (≥18 years) with an initial ICD10 diagnosis of mild depression between 2017 and 2023 were included (N=2131). Nonprogressors were defined as patients whose mild major depressive disorder remained mild for 24 months (N=270). Progressors were defined as patients who developed moderate or severe ICD10 depression within 24 months of the index diagnosis (N=533). Data were stratified and split into (60%) training, (20%) validation, and (20%) test subsets. A heterogeneous feature set spanning demographics, healthcare utilization, socioeconomic indices, diagnostic context, and laboratory measurements were available. Logistic regression utilized elastic net regularization with fivefold cross validation, and random forest hyperparameters were tuned by grid search. XGBoost, CatBoost, and a deep neural network (DNN) were trained with standard learning rate, depth, class weighting, and early stopping. A deterministic top model selection framework applied prespecified thresholds of sensitivity at least 0.70 and AUC at least 0.70, and composite rankings integrated accuracy, sensitivity, specificity, and the overfitting gap. Results: The analytic cohort included 803 patients with complete two-year follow-up. Under the selection criteria, the DNN failed to meet the AUC threshold (0.671) and was excluded. Among the remaining models, XGBoost achieved the top composite score (accuracy = 0.72; AUC = 0.776; sensitivity = 0.77; specificity = 0.63; overfit gap = 0.112). Logistic regression ranked second (accuracy = 0.71; AUC = 0.797; sensitivity = 0.79; specificity = 0.61; overfit gap = 0.052), followed by CatBoost and random forest, the latter penalized for overfitting (gap = 0.278). The TinyLlama audit note, generated through a local Hugging Face pipeline, confirmed XGBoost as the most balanced model. Conclusions: Using EHR data from a multi-institutional regional health system, we developed and validated machine-learning models that predicted progression of depression. XGBoost demonstrated the most reliable composite performance. These findings support the feasibility of leveraging socioeconomic and EHR data to predict worsening depression and emphasize the importance of transparent model-selection frameworks for trustworthy clinical artificial intelligence.
- Open Access
- 28 October 2025
- Published by Elsevier in Neurocomputing
- Vol. 661
Ethics of trustworthy AI in healthcare: Challenges, principles, and practical pathways
- 28 October 2025
- Published by Elsevier in Neurocomputing
- Vol. 661
Open AccessEthics of trustworthy AI in healthcare: Challenges, principles, and practical pathways
- Open Access
- 24 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 144-148
OAAgent: Multimodal LLM Agent for Predicting Knee Osteoarthritis Progression
- 24 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 144-148
Open AccessOAAgent: Multimodal LLM Agent for Predicting Knee Osteoarthritis Progression
- Open Access
- 13 May 2025
- Published by MDPI AG in Preprints.org
A Survey on Hallucination in Large Language and Foundation Models
Generative text models, particularly large language models (LLMs) and foundation models, have influenced numerous fields, including high-quality text generation, reasoning, and multimodal synthesis. These models have been widely applied in healthcare, legal analysis, and scientific research. However, where accuracy and reliability are critical, generative text models pose a significant risk due to hallucination, where generated outputs include incorrect factuality, fabricated, or misleading information. In this survey, we present a review of hallucination in generative AI, covering its taxonomy, detection methods, mitigation strategies, and evaluation benchmarks. We first establish a structured taxonomy, distinguishing between intrinsic vs. extrinsic hallucination and factual vs. semantic hallucination, also discussing task-specific variations in areas such as summarization, machine translation, and dialogue generation. Next, we examine state-of-the-art hallucination detection techniques, including uncertainty estimation, retrieval-augmented generation (RAG), self-consistency validation, and internal state monitoring. We further explore mitigation strategies, such as fine-tuning, reinforcement learning from human feedback (RLHF), knowledge injection, adversarial training, and contrastive learning. Additionally, we review key evaluation metrics and benchmarks, including FEVER, TruthfulQA, HALL-E, and Entity-Relationship-Based Hallucination Benchmarks (ERBench), which serve as standardized measures for assessing hallucination severity. Despite notable efforts, hallucination remains an open challenge, necessitating further improvements in real-time detection, multimodal hallucination evaluation, and trustworthiness frameworks. We show critical research gaps including the need for standardized hallucination taxonomies, scalable mitigation techniques, and human-AI hybrid verification methods. Our survey aims to serve as a foundational resource for researchers and practitioners, providing insights into current methodologies and guiding future advancements in trustworthy and explainable generative AI.
- 13 May 2025
- Published by MDPI AG in Preprints.org
Open AccessA Survey on Hallucination in Large Language and Foundation Models
Generative text models, particularly large language models (LLMs) and foundation models, have influenced numerous fields, including high-quality text generation, reasoning, and multimodal synthesis. These models have been widely applied in healthcare, legal analysis, and scientific research. However, where accuracy and reliability are critical, generative text models pose a significant risk due to hallucination, where generated outputs include incorrect factuality, fabricated, or misleading information. In this survey, we present a review of hallucination in generative AI, covering its taxonomy, detection methods, mitigation strategies, and evaluation benchmarks. We first establish a structured taxonomy, distinguishing between intrinsic vs. extrinsic hallucination and factual vs. semantic hallucination, also discussing task-specific variations in areas such as summarization, machine translation, and dialogue generation. Next, we examine state-of-the-art hallucination detection techniques, including uncertainty estimation, retrieval-augmented generation (RAG), self-consistency validation, and internal state monitoring. We further explore mitigation strategies, such as fine-tuning, reinforcement learning from human feedback (RLHF), knowledge injection, adversarial training, and contrastive learning. Additionally, we review key evaluation metrics and benchmarks, including FEVER, TruthfulQA, HALL-E, and Entity-Relationship-Based Hallucination Benchmarks (ERBench), which serve as standardized measures for assessing hallucination severity. Despite notable efforts, hallucination remains an open challenge, necessitating further improvements in real-time detection, multimodal hallucination evaluation, and trustworthiness frameworks. We show critical research gaps including the need for standardized hallucination taxonomies, scalable mitigation techniques, and human-AI hybrid verification methods. Our survey aims to serve as a foundational resource for researchers and practitioners, providing insights into current methodologies and guiding future advancements in trustworthy and explainable generative AI.
- Open Access
Adopting Trustworthy AI for Sleep Disorder Prediction: Deep Time Series Analysis with Temporal Attention Mechanism and Counterfactual Explanations
Sleep disorders have a major impact on both lifestyle and health. Effective sleep disorder prediction from lifestyle and physiological data can provide essential details for early intervention. This research utilizes three deep time series models and facilitates them with explainability approaches for sleep disorder prediction. Specifically, our approach adopts Temporal Convolutional Networks (TCN), Long Short-Term Memory (LSTM) for time series data analysis, and Temporal Fusion Transformer model (TFT). Meanwhile, the temporal attention mechanism and counterfactual explanation with SHapley Additive exPlanations (SHAP) approach are employed to ensure dependable, accurate, and interpretable predictions. Finally, using a large dataset of sleep health measures, our evaluation demonstrates the effect of our method in predicting sleep disorders.
Open AccessAdopting Trustworthy AI for Sleep Disorder Prediction: Deep Time Series Analysis with Temporal Attention Mechanism and Counterfactual Explanations
Sleep disorders have a major impact on both lifestyle and health. Effective sleep disorder prediction from lifestyle and physiological data can provide essential details for early intervention. This research utilizes three deep time series models and facilitates them with explainability approaches for sleep disorder prediction. Specifically, our approach adopts Temporal Convolutional Networks (TCN), Long Short-Term Memory (LSTM) for time series data analysis, and Temporal Fusion Transformer model (TFT). Meanwhile, the temporal attention mechanism and counterfactual explanation with SHapley Additive exPlanations (SHAP) approach are employed to ensure dependable, accurate, and interpretable predictions. Finally, using a large dataset of sleep health measures, our evaluation demonstrates the effect of our method in predicting sleep disorders.
- Open Access
- 12 June 2024
- Published by Association for Computing Machinery (ACM)
- p. 760-765
A Deep Multimodal Representation Learning Framework for Accurate Molecular Properties Prediction
- 12 June 2024
- Published by Association for Computing Machinery (ACM)
- p. 760-765
Open AccessA Deep Multimodal Representation Learning Framework for Accurate Molecular Properties Prediction