TNIS: A Novel Thyroid Nodule Image Segmentation Model Based on Multi-Channel Feature Fusion

Accurate segmentation of thyroid nodules from medical images is crucial for diagnosing and planning treatment in thyroid-related disorders. However, extracting nodules from thyroid images faces several challenges due to the complexity and variability of nodule morphology and imaging characteristics. In this paper, we propose the Thyroid Nodule Image Segmentation (TNIS) network for accurately segmenting thyroid nodules in medical images. TNIS comprises three interconnected modules designed to enhance segmentation precision through multichannel feature fusion. The first module employs a UNet-based module adept at precise nodule area extraction from input images. Subsequently, we design a boundary detection module to delineate the boundaries of thyroid nodules, facilitating finer segmentation. Additionally, we further integrate a texture feature extraction module to capture and leverage intricate textural nuances inherent to thyroid nodules. By synergistically combining these components, TNIS achieves superior segmentation performance, effectively delineating thyroid nodules with high accuracy and fidelity. Experimental evaluations conducted on diverse datasets validate the efficacy or advantages of TNIS by comparing it with state-of-the-art algorithms.