Internal Medicine · Case Diagnosis
Multi-system case analysis with lab results, imaging, and ECG integration.
Subject Overview
30%
of Total Question Bank
Largest discipline
6
Sub-Domains
Clinical·Imaging·Basic Med·Pharmacology·Public Health·Nursing
8
Imaging Modalities
X-ray·CT·MRI·ECG·WSI·Fundus·Dermoscopy·Endoscopy
Clinical medicine is the largest discipline in Langhui AI STEM-Questions, reflecting its unparalleled importance in multimodal AI. The benchmark spans 6 sub-domains with 8 imaging modalities, simulating real-world diagnostic workflows from patient presentation through imaging interpretation to differential diagnosis and treatment planning. Each question is derived from certified medical institution records and validated through tri-level expert review against clinical gold standards.
Sub-Domains & High-Value Difficulty Zones
Clinical Medicine (35%)
Surgery · Operative Planning
Pre-op imaging interpretation, surgical approach decision, anatomical risk assessment.
Emergency Medicine
Acute triage, trauma assessment, poisoning, cardiac emergency protocols.
Medical Imaging (20%)
Chest X-Ray Interpretation
Pneumonia, pneumothorax, cardiomegaly, pleural effusion detection.
Evidence: CheXpert benchmark F1 0.652 for SOTA VLMs.
CT/MRI Cross-Sectional Anatomy
Head CT, abdominal CT, brain MRI with contrast — lesion localization.
Pathology WSI
Whole-slide image diagnosis, tumor grading, IHC interpretation.
Basic Medicine (15%)
Anatomy & Histology
Gross anatomy diagrams, histological slide identification, embryology schematics.
Physiology & Pathophysiology
Mechanism diagrams, feedback loops, cardiac cycle, renal physiology.
Biochemistry & Molecular Biology
Metabolic pathway diagrams, enzyme kinetics, genetic code interpretation.
Pharmacology (15%)
Drug Mechanisms & Interactions
Receptor binding diagrams, dose-response curves, drug interaction networks.
Clinical Pharmacy
Prescription analysis, TDM (therapeutic drug monitoring), ADR identification.
Public Health (8%)
Epidemiology
Outbreak curves, incidence/prevalence calculations, case-control study design.
Biostatistics
Survival curves, meta-analysis forest plots, ROC curve interpretation.
Nursing (7%)
Clinical Nursing Assessment
Vital sign interpretation, nursing diagnosis, care plan formulation.
Specialized Nursing
ICU monitoring, perioperative care, wound management.
Difficulty Distribution
| Level | Stage | Share | Typical Question Types |
|---|---|---|---|
| L1 Pre-Clinical | Basic sciences (anatomy, physiology, biochemistry) | 10% | Structure identification, pathway recall |
| L2 Clinical Foundation | Clinical clerkship level | 55% | Single-modality imaging diagnosis, case analysis |
| L3 Licensing Exam | USMLE Step 2 CK / equivalent | 25% | Multi-modality integration, differential diagnosis |
| L4 Specialist | Residency / board certification | 10% | Complex multi-system cases, rare disease diagnosis |
Sample Questions
STEM-2026-MED-0000001L2 Clinical Foundation
Chest X-Ray · Lobar Pneumonia Identification
Prompt: A 65-year-old male presents with fever (39.2°C), productive cough, and right-sided chest pain. The accompanying chest X-ray (PA view) shows a dense consolidation in the right upper lobe with air bronchogram sign. Identify the most likely diagnosis and the causal pathogen in community-acquired cases.
Answer: Right upper lobe lobar pneumonia. Most common community-acquired pathogen: Streptococcus pneumoniae. The air bronchogram sign confirms alveolar consolidation while airways remain patent.
5-Model Evaluation
GPT-5.1
3/5
Claude Opus 4.6
2/5
Gemini-3.1-Pro
2/5
Qwen3.6-Plus
1/5
DeepSeek-V4
1/5
STEM-2026-MED-0000002L3 Licensing Exam
ECG · Acute Myocardial Infarction
Prompt: A 58-year-old male presents with crushing substernal chest pain radiating to the left arm, diaphoresis, and nausea. The 12-lead ECG shows ST-segment elevation in leads II, III, and aVF with reciprocal ST depression in leads I and aVL. Identify the infarct location and the most likely culprit coronary artery.
Answer: Inferior wall STEMI. Culprit artery: Right coronary artery (RCA) in ~80% of cases (dominant RCA circulation). Leads II/III/aVF reflect the inferior wall; reciprocal changes in I/aVL confirm.
Step-by-Step Reasoning
- ST elevation in II/III/aVF → inferior wall.
- Reciprocal ST depression in I/aVL → confirms inferior MI.
- Inferior wall supplied by RCA (~80%) or LCX (~20%).
- Need right-sided leads (V4R) to confirm RV involvement.
5-Model Evaluation
GPT-5.1
1/5
Claude Opus 4.6
1/5
Gemini-3.1-Pro
2/5
Qwen3.6-Plus
0/5
DeepSeek-V4
0/5
STEM-2026-MED-0000003L4 Specialist
Pathology WSI · Breast Cancer Grading
Prompt: The accompanying whole-slide image (H&E stain, 20× magnification) shows invasive ductal carcinoma of the breast. Evaluate the Nottingham histologic grade based on tubule formation, nuclear pleomorphism, and mitotic count. Recommend IHC markers for molecular subtyping.
Answer: Nottingham Grade 2 (score 3+3+2=8): tubule formation 10-75% (score 3), marked nuclear pleomorphism (score 3), mitotic count 8-12/10 HPF (score 2). Recommended IHC panel: ER, PR, HER2, Ki-67 for molecular subtyping (Luminal A/B, HER2-enriched, Triple-negative).
Step-by-Step Reasoning
- Tubule formation: scattered tubules within desmoplastic stroma → ~30% → score 3.
- Nuclear pleomorphism: marked variation in size/shape, prominent nucleoli → score 3.
- Mitotic count: 8-12 per 10 high-power fields → score 2.
- Total score 8 → Nottingham Grade 2 (moderately differentiated).
- IHC panel: ER/PR (hormone receptors), HER2 (amplification), Ki-67 (proliferation index).
5-Model Evaluation
GPT-5.1
1/5
Claude Opus 4.6
0/5
Gemini-3.1-Pro
0/5
Qwen3.6-Plus
0/5
DeepSeek-V4
0/5
Delivery & API
JSONL Bulk Download: Full fields + image URLs (DICOM/PNG) + 5-model evaluation + reasoning chains.
REST API: GET /DataAssetsAPI/stem-questions/medical. For enterprise API keys, contact lk@langhuiai.com.
FAQ
How is the difficulty threshold enforced?
Every question must achieve ≤40% pass rate across 5 models × 5 passes (GPT-5.1 / Claude Opus 4.6 / Gemini-3.1-Pro / Qwen3.6-Plus / DeepSeek-V4). Human expert accuracy ≥95%, dual-review gold standard.
What sub-domains does the Medical benchmark cover?
6 sub-domains: Clinical Medicine (35%), Medical Imaging (20%), Basic Medicine (15%), Pharmacology (15%), Public Health (8%), Nursing (7%).
What medical image types are included?
Chest X-ray (DICOM), ECG waveforms, pathology whole-slide images (WSI), CT/MRI sequences, fundus photographs, dermoscopy images, and endoscopy video keyframes.
Is the benchmark clinically validated?
Yes. All cases are drawn from certified medical institutions with tri-level expert review. Diagnosis accuracy verified against clinical gold standards.
How does the benchmark compare to MedQA/USMLE?
Unlike text-only MedQA/USMLE, our benchmark is multimodal (image+text). Zero overlap with any public dataset via pHash+n-gram dedup.