Deep-learning project for classifying glomerular features from renal WSI.

Detected glomeruli are extracted as patches and classified with ResNet-18. The labels describe clinically meaningful glomerular characteristics such as location, appearance, and distribution, even when supervision is originally available at WSI level rather than at single-glomerulus level.

Predictions are made at glomerulus level and aggregated back to WSI-level labels using rules provided by clinicians where available. The project therefore combines model training, label propagation, glomerulus-level inference, and patient/slide-level reporting.

---

← Back to projects