I am a Ph.D. Candidate in Computer Science & Health Informatics at Loyola University Chicago, building AI systems at the intersection of Large Language Models and clinical medicine.
My research focuses on extracting structured knowledge from unstructured clinical text — ICU physician notes, EHRs, and clinical narratives — to build AI tools that genuinely support patient care at the bedside and beyond.
- 🔬 88 citations · 14 publications · 2,100+ reads
- 🏥 Graduate Research Assistant · Loyola University Chicago
- 📍 Chicago, IL
- 💬 Ask me about Clinical NLP, LLMs, EHR AI, Digital Twins, SparkNLP Healthcare
|
🤖 Clinical NLP & LLMs Assertion detection, named entity recognition, and clinical concept extraction from unstructured physician notes using large language models and healthcare-specific NLP pipelines (SparkNLP Healthcare, BioBERT). |
🏥 Digital Twins for ICU Fine-tuning LLaMA-3 with LoRA on ICU physician notes (MIMIC-III) for personalized medication management and clinical decision support in the Medical ICU. |
|
📋 Health Informatics & EHR AI Structured information extraction from Electronic Health Records using scalable NLP pipelines. Patient phenotyping, clinical entity linking, and de-identification with SparkNLP Healthcare NER models. |
⚗️ Generative AI in Medicine Domain-specific embeddings (BioBERT, clinical sentence transformers), performance-driven voting frameworks, and context-specific fine-tuning for clinical AI tasks on i2b2 and MIMIC datasets. |
2026 | 📕 A Hybrid Language Framework for Ontology-Based Clinical Concept Extraction Journal of Healthcare Informatics Research · Springer · doi:10.1007/s41666-026-00232-0 · Behnaz Eslami · Dmitriy Dligach · Nazanin Azarvash · Paula de la Pena · Benjamin Strickland · Samie Tootooni -- SparkNLP NER · SentenceBERT · LLaMA3-8B · Mistral-7B · UMLS · SNOMED CT · MIMIC-III Discharge Summaries
2025 | 📘 Toward Digital Twins in the Intensive Care Unit: A Medication Management Case Study Journal of the American Medical Informatics Association (JAMIA) · doi:10.1093/jamia/ocaf127 · Behnaz Eslami · Majid Afshar . Samie Tootooni . Timothy Miller . Matthew Churpek . Yanjun Gao . Dmitriy Dligach -- LLaMA-3 + LoRA fine-tuned on MIMIC-III · BERTScore 0.842 · Medical ICU specialty adaptation
2025 | 📗 A Performance-Based Voting Framework for Assertion Detection in Clinical Notes · Behnaz Eslami · Dmitriy Dligach . Benjamin Strickland · Nazanin Azarvash . Samie Tootooni -- BioBERT · domain-specific embeddings · performance-driven voting mechanism
Clinical NLP Frameworks
Clinical AI & LLMs
Healthcare NLP Tasks
Datasets & Standards
Languages & Databases
Leveraging John Snow Labs' SparkNLP Healthcare library for production-grade clinical NLP — one of the most comprehensive biomedical NLP frameworks available.
|
🔍 Clinical Named Entity Recognition Extracting medical entities — diagnoses, drugs, dosages, procedures, anatomy — from free-text physician notes and discharge summaries using pre-trained healthcare NER models. |
✅ Assertion Status Detection Classifying clinical findings as present, absent, possible, conditional, or hypothetical — a core capability for accurate clinical AI reasoning, benchmarked on i2b2 datasets. |
|
🔗 Clinical Relation Extraction Identifying relationships between medical entities (e.g., drug–dosage, problem–treatment) to build structured knowledge graphs from unstructured EHR text. |
🔒 De-identification & PHI Removal Applying HIPAA-compliant de-identification pipelines to protect patient privacy in research datasets, enabling safe use of clinical corpora like MIMIC-III. |
|
🗂 ICD-10 / SNOMED Coding Automated medical code assignment from clinical text using entity linking to standardized vocabularies (UMLS, SNOMED CT, RxNorm), reducing manual coding burden. |
⚡ Scalable Spark Pipelines Building distributed NLP pipelines over large EHR corpora with Apache Spark, enabling clinical text analysis at scale across hospital systems. |
researcher = {
"name" : "Behnaz Eslami",
"pronouns" : "she / her",
"affiliation" : "Loyola University Chicago",
"department" : ["Computer Science", "Health Informatics & Data Science"],
"research" : ["Clinical NLP", "LLMs", "Digital Twins", "EHR AI"],
"methods" : ["LLaMA-3", "BioBERT", "LoRA", "PEFT", "BERTScore", "ROUGE-L"],
"clinical_nlp": ["SparkNLP Healthcare", "Clinical NER", "Assertion Detection",
"De-identification", "ICD Coding", "Relation Extraction"],
"ontologies" : ["SNOMED CT", "UMLS", "RxNorm", "ICD-10", "HL7 FHIR"],
"datasets" : ["MIMIC-III", "i2b2", "ICU physician notes"],
"languages" : ["Python", "Java", "PHP"],
"databases" : ["MySQL", "PostgreSQL"],
"hobbies" : ["reading", "exercise", "walking"],
"open_to" : "research collaborations & clinical AI projects",
"motto" : "Failure lays the groundwork for something even greater.",
}