EndoSeg: Ovarian Lesion Segmenter Built for Nebius Challenge, But Don't Call It a Diagnostic Tool

EndoSeg is a binary ovarian-lesion segmentation model fine-tuned on MMOTU ultrasound data, deployed via Nebius serverless jobs and ONNX Runtime Web for privacy-preserving browser inference. The UI shows teal overlays and agreement metrics, but the model outputs only per-pixel lesion masks, not diagnostic labels. The developers explicitly state it is not a medical device and cannot diagnose endometriosis.