An OCT-reading model beats ChatGPT-5 at planning coronary stents

AI system

System: CA-GPT large model on an AI-OCT system (COMPARE study)
Model: A DeepSeek model within CA-GPT, according to Machine Heart.
Release: Exact release not established.
Application developer: Tangdu Hospital, Air Force Medical University (Li Yan's team) with Shenzhen Tsinghua University Research Institute (Zhu Rui's team), on a Vivolight Medtech OCT platform
Role: CA-GPT proposed decisions before and after coronary intervention for comparison with expert assessments.
Configuration: Machine Heart reports that CA-GPT combined an open-source DeepSeek model with retrieval from more than one million cardiovascular papers and guidelines.
Inputs and outputs: The system retrieved cardiovascular literature and proposed decisions before and after coronary intervention.
Autonomy: Not assessed
Workflow: A domain model built on DeepSeek with retrieval over cardiovascular literature and guidelines. Deployment on a Vivolight OCT imaging platform with the team's own image-analysis functions. An expert-derived procedural record from one center used as the reference standard. Scoring of decisions on ten pre-specified metrics across pre- and post-PCI phases.
Human oversight: Not assessed
Relation to AI: CA-GPT is the system under test: a large model on an AI-OCT platform producing pre- and post-PCI decisions that were scored against an expert record. Its reported advantage over ChatGPT-5 and junior physicians is the authors' own measurement in a preprint that has not been peer reviewed. The decisions were not acted on for the 96 patients.
AI contribution: CA-GPT is the system under test: a large model on an AI-OCT platform producing pre- and post-PCI decisions that were scored against an expert record. Its reported advantage over ChatGPT-5 and junior physicians is the authors' own measurement in a preprint that has not been peer reviewed. The decisions were not acted on for the 96 patients.

Benefits

Outcome identifiers: Not assessed
Benefit: stent diameter (90.3% vs. 72.2%, P<0.05) and length selection (80.6% vs. 52.8%, P<0.01)
Category: Improved Task Accuracy
Status: Not assessed
Beneficiary groups: Organizations.
Beneficiaries: Not assessed
Reach: Not assessed
Distribution and access: Not assessed
Duration: Not assessed
Expected?: Yes
Costs and harms: Not assessed

Measurement

Studies: Not assessed
Study design: Not assessed
Measurement: Not assessed
Setting and sample: Not assessed
Measures: stent diameter (90.3% vs. 72.2%, P<0.05) and length selection (80.6% vs. 52.8%, P<0.01)
Comparison: Not assessed
Results: stent diameter (90.3% vs. 72.2%, P<0.05) and length selection (80.6% vs. 52.8%, P<0.01)
Additional value from AI: Not assessed
Limitations: Not assessed
Denominators: Not assessed
Materials for replication: Not assessed

Money

Kind of value: Not assessed
Amount: Not assessed
Currency and prices: Not assessed
Period and unit: Not assessed
Realization: Not assessed
Recipients and payers: Not assessed
Gross or net: Not assessed
Included costs: Not assessed
Comparison: Not assessed
Method: Not assessed
Return and payback: Not assessed

People and places

Participants: Developer: Tangdu Hospital, Air Force Medical University (Li Yan's team) with Shenzhen Tsinghua University Research Institute (Zhu Rui's team), on a Vivolight Medtech OCT platform. Operator: Tangdu Hospital, Air Force Medical University (study lead, per Machine Heart).
Jurisdictions: Benefit: China. Developer: China. Operator: China.
Ownership and control: Not assessed
Dependencies: Not assessed
Cross-border connections: Not assessed
Geopolitical implications: Not assessed

Follow-up

Conditions for replication: OCT image quality and lesion mix comparable to the 96-patient cohort. The expert reference record is accepted as correct. Prospective use with clinicians acting on the output, which the study did not test.
Replication attempts: Not assessed
Later findings: Not assessed
Availability: Not assessed
Institutional decision: Not assessed
Grounds: Not assessed
Action and result: Not assessed
Success criterion: Not assessed

Evidence

Source numbers: AIBD-0378-source-0001–0004
Citations: Numbered references below.
Language: Not assessed
Source relationships: Developer, Independent
Original evidence: Not assessed
Access and versions: Not assessed
Supported findings: Not assessed
Evidence links: Not assessed
Evidence strength: Not assessed
Missing evidence: Not assessed

Grades

Assessed benefits: Organizations.
Depth: Not assessed
Reasons: Organizations: A single-center preprint comparing decision agreement. No patient outcome, deployment or event is reported.
Observed or intended: Not assessed
Rubric: Not assessed
Assessor and date: Not assessed
Alternative assessments: Not assessed
Fill: Independently reported

Entry details

Kind: Not assessed
Related entries: Not assessed
Field: Healthcare / interventional cardiology decision support. OCT-guided percutaneous coronary intervention.
Purpose: CA-GPT was built to interpret intravascular OCT images and generate PCI planning and assessment decisions as expert-level support. The reported agreement with expert records is that purpose.
Dates: Date basis unassessed: 2025 (precision unassessed). Updated: 21 October 2026.
Stage: Not assessed
Version: SHA-256: cbb0cf395873…

References

Other identifiers: oecd_aim: 2025-12-13-c088.
Category mappings: Not assessed
Brief: Not assessed
Revisions: 4 revisions. Latest: 21 October 2026.

Sources

Preprint Developer

Fang W, Wang C, Ma W, et al. (2025). COMPARE: Clinical Optimization with Modular Planning and Assessment via RAG-Enhanced AI-OCT: Superior Decision Support for Percutaneous Coronary Intervention Compared to ChatGPT-5 and Junior Operators. arXiv:2512.10702, submitted 11 December 2025.
News report Independent author

机器之心Pro (2025). 临床PK完胜ChatGPT-5！国内团队造出首个OCT影像AI系统. Sina (k.sina.com.cn), 16 December 2025.
News report Independent author

潘秀玮 (2025). 医生"隔空"操刀，"提香机器人"完成心脏电生理介入手术. 新华网浙江频道 (Xinhuanet Zhejiang), 13 December 2025.
Registry label Independent author

OECD AI Incidents and Hazards Monitor (2025). AI Systems Enhance Cardiac Surgery Precision and Decision-Making in China. OECD.AI, 13 December 2025.