---
language: en
tags:
- medical-imaging
- ultrasound
- vision-transformer
- dinov3
- classification
- research
license: other
---

# DINOv3 Large – Thyroid Nodule Classification

This repository provides **fine-tuned weights** of a DINOv3-Large Vision Transformer
for **thyroid nodule classification from ultrasound images**.

⚠️ **Research and educational use only. Not for clinical use.**

---

## Model description

- Backbone: DINOv3-Large (Vision Transformer)
- Task: Binary classification (benign vs malignant)
- Modality: Thyroid ultrasound
- Training: Fine-tuned from Meta AI DINOv3 pretrained weights

This model is a **derivative work** of DINOv3 released by Meta AI.

---

## Intended use

This model is intended for:
- Academic research
- Benchmarking
- Educational purposes

Not intended for:
- Clinical diagnosis
- Medical decision-making
- Commercial deployment

---

## License and attribution

The original DINOv3 model is released by **Meta AI** under the **DINOv3 License**.

Use of these weights must comply with Meta AI’s licensing terms:
https://ai.meta.com/resources/models-and-libraries/dinov3-downloads/

By using this model, you agree to respect the original license conditions.

---

## Training details

- Dataset: TN5000 (thyroid ultrasound dataset)
- Preprocessing: CLAHE, unsharp masking, adaptive cropping
- Loss: Cross-entropy
- Optimizer: AdamW
- Seed: 42

See the accompanying code repository for full details:
https://github.com/AlexThunder01/Thyroid-Nodule-AI

---

## Citation

```bibtex
@misc{thyroid-nodule-ai-2025,
  author = {AlexThunder01},
  title  = {Thyroid Nodule AI: DINOv3 Fine-Tuned Weights},
  year   = {2025},
  howpublished = {Hugging Face model repository},
}