Tobias Elze - research works

MerMED-FM was trained on around 3·3 million images from 53 publicly available, unlabelled datasets, comprising 713 931 chest x-rays, 292 353 CT slices, 389 885 ultrasound frames, 1 017 712 pathology patches, 333 099 CFP images, 176 719 OCT slices, and 401 059 dermatoscopy images.

Time-varying analysis demonstrated that this association was strongest early in follow-up (HR ≈ 2.5 in year 1) and declined steadily, reaching non-significance (HR ≈ 1.0) by year 4.

At year 6, the cumulative incidence of nvAMD was 17.7% in the exposed cohort versus 15.2% in the nonexposed cohort.

Baseline AMD stage moderated the effect of CS, with weaker effects of CS for intermediate AMD (Interaction HR = 0.82, 95% CI, 0.78-0.92).

The rate of endophthalmitis was significantly higher for corticosteroid implants and pegcetacoplan compared to anti-VEGF injections.

We extended our Ci-SSGAN with a GPT-5.2-to-Qwen3-8B teacher-distilled reasoning module, fine-tuning Qwen3-8B on 2,660 de-identified ophthalmology notes using expert-reviewed rationales.

The incidence of endophthalmitis was 0.097% (1 case per 1031 procedures) after vitrectomy for retinal detachment, 0.1% for trabeculectomy (1 case per 1000 procedures), 0.075% for tube shunts (1 case per 1333 procedures), and 0.068% for scleral buckle surgery (1 case per 1459 procedures).

The incidence after cataract surgery was 0.038% (1 case per 2652 procedures), 0.039% (1 case per 2551 procedures) after standalone minimally invasive glaucoma surgery, 0.02% (1 case per 5013 procedures) after goniotomy, and 0.011% for strabismus surgery (1 case per 9091 procedures).

These limitations constrain the accuracy of datasets used to investigate disease pathogenesis and progression and to develop robust artificial intelligence (AI) systems.

To address this challenge, we introduce Ci-SSGAN (Clinically Informed Semi-Supervised Generative Adversarial Network), a novel framework that leverages large-scale unlabeled clinical text to reannotate patient conditions with improved accuracy and equity.