Recognition of Thyroid Nodules Using Hierarchical Temporal Awareness Networking Scanning

Abstract

Contrast-enhanced ultrasonography (CEUS), a preferred imaging technique for thyroid nodule diagnosis, is able to reveal the vascular distribution within a thyroid nodule right away. With the aim of mining pathologically-related enhancing dynamics and creating predictions in one step without taking into account a native diagnostic dependency, a number of learning-based algorithms have recently been created. In clinics, separating benign from malignant nodules is always done before identifying pathological types. In the current study, we introduce a novel hierarchical temporal attention network (HiTAN), which combines dynamic enhanced deep features and nodule identification into the a complex hierarchy.

This method, in particular, orders a two-stage classification job for diagnosing nodules, with diagnostic reliance modelled by Gated Recurrent Units (GRUs). We also create a local-to-global temporal aggregation (LGTA) operator along the hierarchical prediction path to accomplish a full temporal fusion. Local temporal information is defined precisely as recurrent enhancement patterns in the perfusion representation direction identified at the division level.