ThyCLIPNet: A BiomedCLIP-Guided Lightweight Attention-Enhanced DeepLabV3+ Framework for Robust Thyroid Nodule Segmentation

TL;DR

ThyCLIPNet is a lightweight framework for thyroid nodule segmentation, integrating BiomedCLIP-derived biomedical semantic guidance into a multi-scale CNN pipeline. It achieves high accuracy with low computational cost. ThyCLIPNet outperforms existing methods on several datasets.

Why It Matters

Accurate thyroid nodule segmentation is crucial for diagnosis and treatment. ThyCLIPNet's efficiency and accuracy can improve clinical outcomes and reduce computational costs. Its lightweight design makes it suitable for deployment on edge devices or in resource-constrained environments.

Key Findings

• 96.22% dice similarity coefficient on TG3K dataset
• 8.55M parameters and 22.99G FLOPs
• outperforms existing methods on TN3K, DDTI, and PKTN datasets