Complementary Roles of Image Classification and Vessel Segmentation in AI-Based Screening for Retinopathy of Prematurity Plus Disease in a Kenyan Preterm Cohort

2026-07-07 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision & Pattern Recognition, Medical Imaging AI · Depth: Expert, quick

Summary

An AI-based screening system for Retinopathy of Prematurity (ROP) Plus disease was developed and evaluated using a cohort of 121 Kenyan preterm infants, encompassing 237 eyes and 1,635 fundus images. This study investigated the complementary roles of image classification and vessel segmentation for detecting Plus disease, a subjective condition requiring treatment. Eleven distinct AI configurations, including image classifiers, multiple-instance learning, and multi-task segmentation-classification pipelines, were assessed. While RGB classifiers demonstrated high sensitivity, they also led to over-referral. Vessel segmentation proved feasible, achieving a Dice score of 0.533 and specificity of 0.979. Combining these approaches significantly enhanced performance, with a probability ensemble model yielding the best balanced results: 0.692 sensitivity, 0.914 specificity, and 0.803 balanced accuracy, surpassing standalone vision classifiers.

Key takeaway

For AI Scientists developing diagnostic systems for medical imaging in resource-limited regions, you should integrate both image classification and vessel segmentation into your workflows. This combined approach, particularly using a probability ensemble, offers superior balanced performance (0.692 sensitivity, 0.914 specificity) compared to single-method classifiers, effectively reducing over-referral while maintaining high sensitivity. Prioritize multi-site validation for robust deployment.

Key insights

Combining image classification and vessel segmentation improves AI-based ROP Plus detection by balancing sensitivity and specificity.

Principles

• Classifiers excel at sensitive case-finding.
• Segmentation boosts specificity, reducing over-referral.
• Combined AI workflows yield balanced performance.

Method

Evaluated 11 AI configurations for ROP Plus detection, including image classifiers, multiple-instance learning, multi-task segmentation-classification, and segment-then-classify pipelines, using patient-grouped nested cross-validation.

In practice

• Employ OR-based screening for maximum sensitivity.
• Apply AND-based confirmation for highest specificity.
• Use probability ensembles for balanced detection.

Topics

• Retinopathy of Prematurity
• Plus Disease
• Image Classification
• Vessel Segmentation
• AI Screening
• Medical Imaging

Best for: AI Scientist, Research Scientist, Computer Vision Engineer

Related on AIssential

• A Dual Edge Spatial Jacobian Image Graph for Interpretable Diabetic Retinopathy Grading — Interpretable DR grading benefits from graph representations linking spatial lesion-vessel geometry and embedding-biomarker sensitivity.
• Could AI Imaging Spot Pregnancy Health Risks Earlier? — AI can significantly enhance early detection of high-risk pregnancy conditions like PAS and CHD, improving care pathways.
• Trusted Multi-View Deep Learning Classification of Fetal Congenital Heart Disease with Feature-level and Decision-level Fusion — The framework integrates multi-view echocardiographic data with uncertainty handling for robust fetal CHD classification.
• Towards Global AI-Driven Cervical Cancer Screening — A multi-task deep learning model validated on diverse global data outperforms experts in cervical cancer screening.
• Cross-Source Supervision for Bone Infection Segmentation in Dual-Modality PET-CT — Decoupled dual-source learning with PET-CT fusion improves bone infection segmentation by accommodating diverse clinical annotation philosophies.
• AUBADE-syn: a novel deep learning ensemble method for glaucoma detection using synthetic fundus images on imbalanced datasets — AUBADE-syn improves glaucoma detection on imbalanced datasets via synthetic image generation and ensemble learning.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.