Three-stage pipeline for automated dental analysis on panoramic X-rays quadrant detection, tooth enumeration, and pathology classification (caries, deep caries, periapical lesions, impacted teeth) using the DENTEX dataset.

Two-stage intraoral caries screening: fine-tuned YOLO detector + MedGemma 1.5 for per-lesion reasoning. Beats the published benchmark on a patient-safe test split.

PyTorch Lightning implementation of a 5-class dental pathology segmentation system using a U-Net with a pre-trained ResNet34 backbone, joint Dice-CE loss, and power-law smoothed class weighting.