Physician researcher / medical AI connector

Nariman Naderi

I work at the intersection of clinical reasoning, research design, technical implementation, and healthcare workflow reality.

My current work focuses on medical AI in gastroenterology, LLM reliability, clinical guideline synthesis, vision-language models, and imaging dataset quality, with a longer-term direction toward radiology AI and practical medical computer vision.

View publications See project evidence

About

A clinical bridge for practical medical AI

My strongest role is connecting the clinical problem, the research frame, the implementation path, and the real-world failure modes.

I am a physician researcher based at the Gastroenterology and Liver Research Center, Taleghani Hospital, working on AI in gastroenterology and broader medical AI. I am a general physician with broad clinical exposure and a technical direction shaped by Python, machine learning, deep learning, computer vision, and large language models.

I am most interested in systems that are useful under clinical constraints: reliable enough to study seriously, transparent enough to evaluate, and practical enough to fit real healthcare workflows.

Clinical reasoning Research design Technical implementation Workflow reality

Research

Current themes

Research-first proof: selected themes where clinical relevance, reliability, and evaluation matter more than benchmark appearance alone.
01

LLM confidence and uncertainty

Evaluating how medical large language models express confidence, uncertainty, and calibration in clinical reasoning tasks.

Reliability / calibration
02

Clinical guideline RAG

Using retrieval-augmented generation to synthesize overlapping or conflicting clinical guidelines into structured outputs.

Guidelines / synthesis
03

Vision-language models

Studying when general-purpose vision-language models can compete with trained models in medical image tasks.

Endoscopy / VLMs
04

Imaging dataset quality

Assessing medical imaging datasets for clinical relevance, bias, redundancy, and real-world applicability.

Dataset realism / bias

Publications

Selected work

Google Scholar
Journal article / first author

Across generations, sizes, and types, large language models poorly report self-confidence in gastroenterology clinical reasoning tasks.

npj Gut and Liver. First author; equal contribution with SAA Safavi-Naini.

DOI
Journal article / RAG

Using large language models to integrate international IBD guidelines: A retrieval-augmented generation approach.

Colorectal Disease. Contributed author.

DOI
Journal article / vision-language models

Vision language models versus machine learning models performance on polyp detection and classification in colonoscopy images.

Scientific Reports. Contributed author.

DOI
Workshop paper / first author

Evaluating prompt engineering techniques for accuracy and confidence elicitation in medical LLMs.

EXTRAAMAS / Lecture Notes in Computer Science.

DOI
Preprint / dataset review

State of abdominal CT datasets: A critical review of bias, clinical relevance, and real-world applicability.

arXiv. Contributed author.

arXiv

Projects

Builder evidence

Projects are presented as evidence of direction and capability, not as claims of finished clinical products.
Medical imaging AI

Perifistaid MRI segmentation work

Perianal fistula MRI segmentation validation and manuscript work, including segmentation-oriented imaging workflows and research evaluation.

Proves: segmentation validation, MRI workflow thinking, manuscript-level research discipline.

Local AI app

Question-generation workflow

Planning and building a local-first AI question-generation and review workflow for structured educational content.

Proves: practical app design, local-first AI thinking, review workflow structure.

LLM evaluation

Medical LLM confidence studies

Designed and evaluated model comparisons, prompt variants, confidence elicitation, and uncertainty analysis for medical reasoning tasks.

Proves: model evaluation, uncertainty framing, clinical reasoning experiments.

Guideline synthesis

IBD guideline RAG workflow

Contributed to a retrieval-augmented generation approach for integrating international inflammatory bowel disease guidelines.

Proves: RAG concepts, clinical guideline structure, expert-aligned synthesis.

Technical profile

Practical tools for research and prototypes

Strongest areas

  • Python scripts, data workflows, and research utilities
  • Medical LLM experiments, RAG, prompting, and evaluation
  • CSV-heavy data cleaning and analysis workflows
  • Clinical AI problem framing and research translation

Comfortable areas

  • Deep learning and computer-vision workflows
  • CNN/U-Net-style segmentation pipelines
  • Flask, FastAPI, Streamlit, and lightweight apps
  • Linux, Windows, remote terminals, and GPU environments

Growth areas

  • More polished public repositories with reproducible READMEs
  • Advanced Git collaboration patterns
  • Deeper concurrency, multiprocessing, and performance profiling

Insights

Short notes on medical AI systems

Brief field notes that connect research questions, implementation choices, and clinical constraints.
Computer vision

Medical imaging AI beyond the model

Why segmentation work depends on data structure, preprocessing, evaluation, and clinical meaning as much as architecture choice.

Contact

Collaboration around medical AI, LLM reliability, and imaging AI

The best starting points are research collaboration, medical AI evaluation, clinical guideline synthesis, computer vision, and portfolio-level technical projects.
GitHub Google Scholar
LinkedIn X / Twitter Email