Thyroid Nodule Ultrasound Images for Benign-Malignant Classification

Fine-tuned weights of a DINOv3-Large Vision Transformer model are provided for binary classification of thyroid nodules from ultrasound images. The model was created by alexThunder01 and fine-tuned from Meta AI's pretrained DINOv3 weights. The repository was last updated in January 2026.

Use Cases

• Classify thyroid ultrasound images as benign or malignant using the fine-tuned DINOv3-Large model.
• Benchmark vision transformer performance on a binary medical imaging task using the provided model weights.
• Transfer learning for related medical image classification tasks by leveraging the fine-tuned feature extractor.
• Research into model interpretability by analyzing the transformer's attention maps on ultrasound image patches.

Strengths

• Model is fine-tuned from the DINOv3-Large backbone, a state-of-the-art vision foundation model.
• Task is clearly defined as binary classification for a specific medical modality (thyroid ultrasound).

Limitations

• The underlying dataset size, composition, and performance metrics for the fine-tuned model are not provided.
• Potential for bias exists as the source data's geographic and demographic coverage is unknown.

Provenance

Source

alexThunder01 on Hugging Face.

Collection Method

Model was fine-tuned from Meta AI's DINOv3 pretrained weights on a thyroid ultrasound dataset (specifics unknown).

Freshness

Repository was updated on 2026-01-24.