Maintenance Notice

Due to necessary scheduled maintenance, the JMIR Publications website will be unavailable from Wednesday, July 01, 2020 at 8:00 PM to 10:00 PM EST. We apologize in advance for any inconvenience this may cause you.

Who will be affected?

Currently submitted to: JMIR AI

Date Submitted: Oct 7, 2026

Warning: This is an author submission that is not peer-reviewed or edited. Preprints - unless they show as "accepted" - should not be relied on to guide clinical practice or health-related behavior and should not be reported in news media as established information.

Automated Kaplan-Meier Survival Curves Digitalization and Individual Patient Data Reconstruction Using GenAI and Computer Vision: System Development and Validation

  • Ying Li; 
  • Kristin Qi; 
  • Liang-Chin Huang; 
  • Jianguang Zhuang; 
  • Siwei Wang; 
  • Long He; 
  • Dylan Cote; 
  • Matthew Cmiel; 
  • Majid Rastegar-Mojarad; 
  • Jingcheng Du; 
  • Chenguang Wang; 
  • Bo Gao; 
  • Patricia Guyot; 
  • Yingxin Xu

ABSTRACT

Background:

Accurate digitization of Kaplan-Meier (KM) curves and reconstruction of individual patient data (IPD) from published graphs are important for health technology assessment and evidence synthesis but remain technically challenging.

Objective:

To develop and validate a human-in-the-loop system that uses generative artificial intelligence and computer vision to digitize published KM curves and reconstruct IPD for downstream health technology assessment and evidence synthesis.

Methods:

We developed a KM-Vision system that integrates figure quality control screening, metadata extraction using a large language model (LLM), computer vision (CV)-based curve tracing, IPD reconstruction using an enhanced Guyot-based algorithm, and an accuracy assessment panel. We evaluated performance using published KM figures, synthetic KM figures with known ground truth, expert-digitized figures, and clinical trial patient-level records. On these datasets, we compared KM-Vision with general-purpose LLM models (GPT-4o and Claude 3.7), KM-GPT (a published automated KM digitization pipeline), and expert manual digitization.

Results:

KM-Vision outperformed general-purpose LLMs and KM-GPT on published figures and synthetic data. For published dual-arm KM figures, mean median survival time (MST) deviation was 0.092 months and mean hazard ratio (HR) error was 0.017, substantially lower than GPT-4o (18.910 and 0.423) and Claude 3.7 (24.090 and 0.277). Performance was also comparable to expert digitization, with similar MST deviations (0.384 and 0.029 vs 0.076 and 0.035) and HR errors (0.071 and 0.022 vs 0.029 and 0.016) across published figures and trial data.

Conclusions:

KM-Vision, a high-accuracy, verifiable LLM-CV pipeline, enables expert-level reconstruction of IPD from published KM curves for use in cost-effectiveness analyses (CEA) and indirect treatment comparisons (ITC).


 Citation

Please cite as:

Li Y, Qi K, Huang LC, Zhuang J, Wang S, He L, Cote D, Cmiel M, Rastegar-Mojarad M, Du J, Wang C, Gao B, Guyot P, Xu Y

Automated Kaplan-Meier Survival Curves Digitalization and Individual Patient Data Reconstruction Using GenAI and Computer Vision: System Development and Validation

JMIR Preprints. 07/10/2026:113843

DOI: 10.2196/preprints.113843

URL: https://preprints.jmir.org/preprint/113843

Download PDF


Request queued. Please wait while the file is being generated. It may take some time.

© The authors. All rights reserved. This is a privileged document currently under peer-review/community review (or an accepted/rejected manuscript). Authors have provided JMIR Publications with an exclusive license to publish this preprint on it's website for review and ahead-of-print citation purposes only. While the final peer-reviewed paper may be licensed under a cc-by license on publication, at this stage authors and publisher expressively prohibit redistribution of this draft paper other than for review purposes.