Talk to a Concierge  →
Home  ›  Resources  ›  Peking Union Medical College Hospital  ›  AMD treatment prediction model
Medical AI · English Editorial Summary

KongMing model forecasts how macular degeneration patients may respond to anti-VEGF therapy

A Lancet Digital Health study by PUMCH ophthalmologists and Shanghai Jiao Tong University reports a deep learning system that predicts visual and anatomical outcomes of anti-VEGF treatment for neovascular age-related macular degeneration, built on data from 18 tertiary hospitals across 12 Chinese provinces.

Original report: 13 April 2026English summary: 18 April 2026Source: PUMCH official websiteEdited by Wellward China
Peking Union Medical College HospitalBeijingEyes & VisionAge-related macular degenerationAnti-VEGFArtificial intelligenceOCT
Source and editorial note: This is an original English summary based on PUMCH’s Chinese-language report, published 13 April 2026, about a study published in The Lancet Digital Health on 26 March 2026. It is not a full translation or an independent clinical assessment.
What was developed

A three-in-one prediction system for wet AMD treatment

The “KongMing” model is a deep learning system built on a lesion-aware Transformer multi-task architecture. It was developed by PUMCH ophthalmologist Chen Youxin’s team with Sheng Bin’s team at Shanghai Jiao Tong University, using data from 18 tertiary hospitals across 12 Chinese provinces. The system performs three tasks: classifying whether visual function will improve, stabilise or worsen; predicting future visual acuity values; and generating expected post-treatment OCT scans. It covers three decision time points — after a single injection, after the three loading doses, and after one year of a 3+PRN regimen.

What the study reported

Reported performance — retrospective validation, not routine deployment

According to the report, KongMing’s vision-change predictions were more accurate than clinicians’, its visual acuity predictions had small errors, and its generated OCT images closely resembled real follow-up scans. It outperformed common baseline models across evaluation metrics. The model also flagged prognostic imaging features consistent with prior literature: choroidal neovascularisation, old scarring and intraretinal fluid were associated with poorer outcomes, while subretinal fluid was associated with better ones.

What it does not do

KongMing is decision support, not an autonomous decision-maker — treatment choices remain with ophthalmologists and patients. The published account describes model development and validation on collected data; it does not describe prospective trials of whether model-guided care leads to better vision outcomes, nor routine clinical deployment. Predictions are probabilities, not guarantees for any individual eye.

Uniform editorial cover image used across Wellward China resource summaries (placeholder)

What comes next

The report suggests the model could help doctors and patients set expectations, adjust plans early when predicted response is poor, and avoid ineffective treatment. Prospective validation and integration into care pathways are the natural next steps, though the report does not state a timeline.

Relevance for patients

Neovascular age-related macular degeneration is a leading cause of vision loss in older adults, and China’s ageing population makes treatment capacity and personalisation increasingly important. This falls under Wellward China’s Eyes & Vision health area. This page is general information, not medical advice.