Abstract:Accurate segmentation of the thyroid gland in ultrasound images is an essential initial step in distinguishing between benign and malignant nodules, thus facilitating early diagnosis. Most existing deep learning-based methods to segment thyroid nodules are learned from only a single view or two views, which limits the performance of segmenting nodules at different scales in complex ultrasound scanning environments. To address this limitation, this study proposes a multi-view learning model, abbreviated as MLMSeg. First, a deep convolutional neural network is introduced to encode the features of the local view. Second, a multi-channel transformer module is designed to capture long-range dependency correlations of global view between different nodules. Third, there are semantic relationships of structural view between features of different layers. For example, low-level features and high-level features are endowed with hidden relationships in the feature space. To this end, a cross-layer graph convolutional module is proposed to adaptively learn the correlations of high-level and low-level features by constructing graphs across different layers. In addition, in the view fusion, a channel-aware graph attention block is devised to fuse the features from the aforementioned views for accurate segmentation of thyroid nodules. To demonstrate the effectiveness of the proposed method, extensive comparative experiments were conducted with 14 baseline methods. MLMSeg achieved higher Dice coefficients (92.10% and 83.84%) and Intersection over Union scores (86.60% and 73.52%) on two different thyroid datasets. The exceptional segmentation capability of MLMSeg for thyroid nodules can greatly assist in localizing thyroid nodules and facilitating more precise measurements of their transverse and longitudinal diameters, which is of significant clinical relevance for the diagnosis of thyroid nodules.

SSE: Scale-adaptive Soft Erase Weakly Supervised Segmentation Network for Thyroid Ultrasound Images

Mega-Trials: Methodological Issues and Clinical Implications

Segmentation of Thyroid Glands and Nodules in Ultrasound Images Using the Improved U-Net Architecture.

Real-time reliable semantic segmentation of thyroid nodules in ultrasound images

A Weakly Supervised Deep Active Contour Model for Nodule Segmentation in Thyroid Ultrasound Images

A devised thyroid segmentation with multi-stage modification based on Super-pixel U-Net under insufficient data

Utilization of Neuraminic Acid Receptors by Mycoplasmas

Comparative efficiency of primary light conversion in photosynthetic bacteria Rhodospirillum rubrum and Rhodopseudomonas viridis.

U2F-GAN: Weakly Supervised Super-pixel Segmentation in Thyroid Ultrasound Images

Ultrasound Image Segmentation of Thyroid Nodule via Latent Semantic Feature Co-Registration

AMSeg: A Novel Adversarial Architecture based Multi-scale Fusion Framework for Thyroid Nodule Segmentation

Beyond Point Annotation: A Weakly Supervised Network Guided by Multi-Level Labels Generated from Four-Point Annotation for Thyroid Nodule Segmentation in Ultrasound Image

MLMSeg: A multi-view learning model for ultrasound thyroid nodule segmentation

[Idiopathic portal hypertension with splenic infarct. An unreported complication].

A segmentation-based algorithm for classification of benign and malignancy Thyroid nodules with multi-feature information

Federated learning via multi-attention guided UNet for thyroid nodule segmentation of ultrasound images

Ultrasound Image-Based Thyroid Nodule Automatic Segmentation Using Convolutional Neural Networks

Automated Segmentation of Thyroid Nodule, Gland, and Cystic Components From Ultrasound Images Using Deep Learning

3D Thyroid Segmentation in CT Using Self-Attention Convolutional Neural Network

Ultrasound Nodule Segmentation Using Asymmetric Learning with Simple Clinical Annotation

The Eighth Meryon Society Lecture read at Worcester College, Oxford on 2 July, 2004