HD-DinoMoE: A Class-Aware Hierarchical Dual Mixture-of-Experts Network for Scleral Anomaly Segmentation in Complex Acquisition Scenarios
Researchers have developed HD-DinoMoE, a novel network designed for segmenting scleral anomalies in ocular inspection images. This system aims to bring objectivity and quantification to Traditional Chinese Medicine's ocular diagnostics. The network employs a class-aware hierarchical dual mixture-of-experts approach to handle variations in image acquisition, anomaly types, and specular reflections, achieving competitive segmentation results on a new benchmark dataset. AI
IMPACT This model could enable more objective and scalable ocular diagnostics in Traditional Chinese Medicine.