ThyroidAgent 0.1.1

DINOv3 + UNet for single-task segmentation and multi-task models (segmentation / malignancy classification / TIRADS), with dataset and training utilities

ThyroidAgent

ThyroidAgent is a PyTorch-based library that provides DINOv3 + UNet models for thyroid ultrasound image analysis, including both single-task segmentation and multi-task learning (segmentation + malignancy classification + TIRADS).

Features

• DINOv3_S_UNet: single-task tumor segmentation.
• DINOv3_S_UNet_MULTITASK: joint segmentation, malignancy classification, and TIRADS prediction.
• Dataset utilities: FullDataset and MultiTaskDataset for training/validation pipelines.
• Metrics and loss: ready-to-use evaluation metrics and loss functions for medical image segmentation and multi-task learning.

Installation

From PyPI (after you publish):

pip install ThyroidAgent