Quantum‐Guided Textural–Spatial Fusion Network for Thyroid Nodule Classification in Ultrasound Imaging

Abstract

Thyroid cancer is among the most common malignancies, and ultrasound is the primary imaging modality due to its safety and wide availability. Accurate diagnosis remains difficult because of low contrast, acoustic shadowing, and subtle visual differences between benign and malignant nodules. These factors increase observer variability and lead to missed or incorrect diagnoses. Prior deep learning methods address this problem only partially by selectively focusing on spatial information and failing to capture long‐range dependencies. Other approaches rely on handcrafted texture descriptors and overlook critical grayscale patterns in ultrasound. This paper presents a Quantum‐guided Textural–Spatial Fusion Network (QTexS‐Net) for thyroid nodule classification in ultrasound imaging. The proposed framework jointly learns complementary spatial and textural representations. A spatial‐aware autoencoder captures local structures and global contextual dependencies, while a texture‐aware branch preserves fine‐grained grayscale characteristics relevant to malignancy. A quantum‐inspired fusion mechanism models interactions between spatial and textural features, leading to more discriminative representations and improved training efficiency. Experiments on three public benchmark datasets show that QTexS‐Net consistently outperforms existing methods. Ablation studies and cross‐dataset evaluations confirm the contribution of each component. Explainability analyses further highlight clinically meaningful regions, supporting reliable, interpretable decision‐making in thyroid nodule assessment and providing transparent decision support for clinicians.