Inha University · Digital Medicine

From Too Much Information to Translational Medical Intelligence

Starting with clinical questions, we study digital biomarkers from medical images, biosignals and electronic medical records (EMR), and develop AI to support clinical decisions and workflows.

Inha University · Digital Medicine

TMI-lab

Translational
Medical Intelligence

Starting with clinical questions, we study digital biomarkers from medical images, biosignals and electronic medical records (EMR), and develop AI to support clinical decisions and workflows.

Department of Digital Medicine, Inha University College of Medicine

TMI-lab dragon and goose logo

Our mission

From information
to intelligence

TMI-lab studies how data and AI can address problems in clinical practice. Our motto, “From Too Much Information to Translational Medical Intelligence”, expresses our aim to turn abundant medical information into knowledge and technology that support care. Both phrases share the initials TMI, linking the lab’s name with its research direction.

Better predictions should lead to better clinical decisions. We connect imaging, biosignals and clinical records to address workflow bottlenecks, anticipate treatment outcomes and build evidence for individual care.

Clinical questions

Start with a clinical question

Where do decisions stall, and what information is missing? We define the question and evaluation criteria with clinicians.

Data & methods

Interpret data in context

We design methods around anatomy, longitudinal change and the clinical context of each problem.

Clinical evaluation

Connect performance to clinical value

We evaluate accuracy, uncertainty and performance on new data, then study how the results can inform care.

AVIEW NeuroCAD

Flagging suspected hemorrhage on brain CT for priority review.

Translation

From research
to clinical impact

Coreline Soft · 2024

AVIEW NeuroCAD

Earlier review for suspected brain hemorrhage

Analyzes suspected hemorrhage and its volume in brain CT to prioritize review. Image-based triage supports faster clinical decisions when timely treatment matters.

Brain CT analysis interface of Coreline Soft AVIEW NeuroCAD
Product image: Coreline Soft · official product page

Adoption and evidence

More than 50,000 clinical uses

A June 2026 report records more than 50,000 cumulative uses as of April 2026. This measures product use, not the number of hospitals.

Hans Economy · 26 Jun 2026

Supporting priority review in emergency care

Pohang Semyung Christian Hospital has adopted NeuroCAD in emergency care. Alerts and quantitative findings from brain CT support clinicians’ initial review.

Daily Medi · 7 Apr 2026

Additional context: use in over 100 emergency departments is lab-provided information from September 2026, a separate measure from the reported cumulative use count.

Product features and regulatory information: Coreline Soft

Medical Image Analysis · 2022

AVIEW NeuroCAD brain CT analysis
Coreline Soft · AVIEW NeuroCAD

The opening illustration is an AI-generated concept of priority review. The image above shows the actual product.

A portfolio of translation

Anatomical landmarks in a Grashey X-ray
GreyNet · Shoulder landmarks for image quality assessment

GreyNet

Itphy

Quality checks at acquisition

Evaluates Grashey X-ray positioning and quality to request an immediate retake from the radiographer when needed, enabling quality checks before the patient leaves the imaging room.

Landmarks in weight-bearing foot X-rays
FlatNet · Weight-bearing foot landmarks for flatfoot assessment

FlatNet

Promedius

More objective flatfoot assessment

Detects landmarks on weight-bearing lateral foot X-rays. A comparative study found lower landmark errors than an orthopedic surgeon, supporting more consistent measurements for flatfoot assessment.

Read the comparative study
Retinal vessel segmentation comparison
ProRetina · Bayesian vessel segmentation comparison

ProRetina

Promedius

Segmentation with quality assessment

Uses Bayesian U-Net uncertainty to improve retinal vessel segmentation while assessing image quality, bringing the analysis and its reliability into the same model.

Shoulder Grashey radiograph with anatomical landmarks from readers and an AI model
GreyNet · Shoulder landmarks for image quality assessment

Image quality

Check image quality
before the patient leaves

GreyNet detects landmarks in Grashey X-rays to check positioning and quality, requesting an immediate retake from the radiographer when needed. Its purpose is to connect acquisition and quality assurance in one workflow.

Read the study

Clinical workflow

How can we spot changes that are easy to miss in radiology reports?

We compare earlier and follow-up reports to find inconsistencies that need a second look.

Illustrative CT comparison

Prior examCT
Synthetic abdominal CT comparison: the same liver focus appears larger in the right follow-up than in the left prior exam. These are not patient scans or measured results.
Liver focus10 mm
ReportBaseline
Follow-upCT
Synthetic abdominal CT comparison: the same liver focus appears larger in the right follow-up than in the left prior exam. These are not patient scans or measured results.
Liver focus16 mm
Report“No change”
LLM report comparison → Discrepancy → Clinical review

Research

Research goal 1. Clinical workflow

From an AI output to the next clinical action

From acquisition and interpretation to record review, we study how AI can surface overlooked signals and repeated checks at the point where clinicians need them.

Methods
Image quality · Longitudinal reports · LLM error detection
Clinical relevance
Surface information that warrants review within the clinical workflow, reducing repeated checks.

Illustrative CT comparison

Prior examCT
Synthetic abdominal CT comparison: the same liver focus appears larger in the right follow-up than in the left prior exam. These are not patient scans or measured results.
Liver focus10 mm
ReportBaseline
Follow-upCT
Synthetic abdominal CT comparison: the same liver focus appears larger in the right follow-up than in the left prior exam. These are not patient scans or measured results.
Liver focus16 mm
Report“No change”
LLM report comparison → Discrepancy → Clinical review

Reading radiology reports across time

This study evaluates errors between prior and follow-up reports. Error-specific performance and false positives are assessed before clinical use.

Study design comparing an LLM and radiologists across six error types injected into 100 longitudinal report pairs
Longitudinal report error detection · Study design

Digital twins

Can we predict how an individual patient will change after treatment?

We use preoperative images and clinical data to study how each patient’s anatomy may change after surgery.

Digital twin research concept comparing two hypothetical surgical outcomes from a patient-specific anatomical model.

Research

Research goal 2. Digital twins

Patient-specific models of what comes next

We model anatomy and its change over time. Building on growth-prediction research, we are extending this work toward surgical digital twins that predict postoperative states from preoperative images and clinical data.

Methods
Anatomical landmarks · Longitudinal models · Outcome prediction
Clinical relevance
Study treatment scenarios to build predictive evidence for surgical planning and patient discussions.

From anatomical change to surgical planning

Medical images compare baseline, observed follow-up and AI-predicted landmarks. This growth and orthodontic-treatment research is a foundation for extending prediction to surgical outcomes and treatment scenarios.

Left: baseline cephalogram overlaid with baseline, observed follow-up and predicted landmarks. Right: follow-up cephalogram comparing observed and predicted landmarks. This is growth and orthodontic-treatment research, not a surgical outcome comparison.
Growth & orthodontic treatment · Blue: baseline / Green: follow-up / Red: AI prediction
Four box plots of mean growth-prediction error in millimeters by treatment duration, appliance type, baseline age and follow-up age
Growth-prediction error · By duration, appliance and age

Precision medicine

How do risk and treatment response vary among patients with the same diagnosis?

We look for digital biomarkers of risk and treatment response in images, biosignals and electronic medical records (EMR).

Multimodal research concept: retinal images, biosignals and medical images converge for interpretation and individual clinical decisions.

Research

Research goal 3. Precision medicine

Beyond a shared diagnosis, evidence for the individual

We analyze medical images, biosignals and electronic medical records (EMR) to identify digital biomarkers and study individual risk and treatment response. We also assess the reliability of measurements and predictions.

Methods
Bayesian uncertainty · Quantitative imaging · Biosignals
Clinical relevance
Interpret measurements alongside their reliability as a foundation for individual risk and treatment-response research.

Precise measurements, interpretable evidence

Retinal vessel segmentation and ECG analysis are foundations for precision medicine. We study how to connect these methods to early prediction and treatment decisions.

Retinal images and vessel segmentation maps comparing methods with and without Bayesian modeling
ProRetina · Bayesian vessel segmentation comparison
ECG attention maps showing the waveform regions emphasized by DCAM
DCAM · Regions emphasized in ECG analysis
Explore the DCAM architecture

Research output

Google Scholar Keewon Shin · Research profile

Checked

Total citations
1,120
h-index
15
i10-index
18

Citations by year

Citations
* 2026: partial year · Citations through the checked date. Metrics cover all time.

14 publications

Patents

Patents & applications

Earlier industrial research patents

People

Professor Keewon Shin
Principal Investigator

Translational medical AI

Keewon Shin

Assistant Professor · PhD

Department of Digital Medicine, Inha University College of Medicine · Deputy Director, AI Big Data Center, Inha University Hospital

Translational research brings advances in AI into clinical practice. His work connects medical imaging, biosignals, and multimodal AI with clinical workflows, surgical outcome prediction, and evidence-based treatment.

Connecting medical imaging, biosignals and multimodal AI with clinical workflows and evidence-based treatment.

Education

2023PhD · Biomedical EngineeringUniversity of Ulsan

MS / BS · Mechanical Engineering, Hanyang University

Academic service

2024–MICCAIArea/Program Chair

2025–Korean Society of AI in MedicineScientific Committee Member

People

People

Professor Keewon Shin

Keewon Shin

Assistant Professor · PhD

Department of Digital Medicine, Inha University College of Medicine · Deputy Director, AI Big Data Center, Inha University Hospital

Translational research brings advances in AI into clinical practice. His work connects medical imaging, biosignals, and multimodal AI with clinical workflows, surgical outcome prediction, and evidence-based treatment.

Education · Career & academic service

Education

  • 2023 · PhD, Biomedical Engineering, University of Ulsan
  • 2013 · MS, Mechanical Engineering, Hanyang University
  • 2011 · BS, Mechanical Engineering, Hanyang University

Career & academic service

  • 2025–present · Scientific Committee Member, Korean Society of Artificial Intelligence in Medicine
  • 2024–present · Area/Program Chair, MICCAI
  • 2026–present · Assistant Professor, Inha University College of Medicine
  • 2025–2026 · Research Professor, University of Ulsan College of Medicine
  • 2023–2025 · Research Professor, Korea University College of Medicine
  • 2018–2023 · Researcher and Postdoctoral Fellow, Asan Medical Center
  • 2013–2018 · Researcher, Hyundai Motor Company
Official Inha University profile

Research activity

Research profile

Medical Multi-Modal AI · Clinical AI Implementation

Citations
1,120
h-index
15
i10-index
18

Google Scholar · All time · Checked

Citations by year

  1. 34
  2. 86
  3. 128
  4. 171
  5. 202
  6. 278
  7. 200
Google Scholar · Through November 15, 2026 · current year is partial

Research collaboration

Collaborative research in medical imaging, biosignals and multimodal AI

kevinkwshin@inha.ac.kr

Room 318, 60th Anniversary Memorial Hall, Inha University

Department of Digital Medicine, Inha University College of Medicine

Contact

Research collaboration
& inquiries

For medical AI collaboration and research inquiries, please contact us at the work email below.

Email
kevinkwshin@inha.ac.kr
Office
Room 318, 60th Anniversary Memorial Hall, Inha University