New segmentation models show promise for tracking eye disease lesions

Researchers have developed and evaluated four lesion-segmentation pipelines, including 2D and 3D variants for AMD and DME, achieving Dice scores between 0.76 and 0.82 on an in-domain validation set. These pipelines demonstrated strong volumetric and surface calibration, with correlations of 0.97 or higher. The study also introduced a full-volume, calibration-aware adoption standard to identify mechanisms missed by slice-level evaluations, finding that ensemble composition consistently improved performance. When tested on an external clinical cohort (OLIVES) using proxy metrics like biomarker AUROC and longitudinal concordance, the models showed promise in tracking clinical biomarkers outside the training distribution, suggesting potential as a clinical tool for automated lesion-burden tracking. AI

IMPACT This research could lead to improved automated tools for diagnosing and monitoring eye diseases like AMD and DME.

RANK_REASON The cluster contains a research paper detailing new methods and evaluations for medical image segmentation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• AMD
• arXiv
• AUROC
• Dice scores
• DME
• OLIVES

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →