Xu 2024a - Revision history

Rimni at 12:27, 6 June 2024

2024-06-06T12:27:24Z

Xupeng336 at 12:22, 6 June 2024

2024-06-06T12:22:42Z

Rimni at 12:00, 6 June 2024

2024-06-06T12:00:53Z

Rimni: /* 3.1.3 Tokenized MLP stage */

2024-06-06T11:58:27Z

‎3.1.3 Tokenized MLP stage

Rimni: /* 1. Introduction */

2024-06-06T11:55:37Z

‎1. Introduction

Rimni at 11:50, 6 June 2024

2024-06-06T11:50:42Z

Rimni: /* 6. Conclusions */

2024-06-06T11:50:26Z

‎6. Conclusions

Rimni at 11:48, 6 June 2024

2024-06-06T11:48:59Z

Rimni at 11:17, 6 June 2024

2024-06-06T11:17:06Z

Show changes

Rimni at 11:13, 6 June 2024

2024-06-06T11:13:22Z

@@ Line 257: / Line 257: @@
 |}
-In Eq. (11), <math>\epsilon</math> represents a smoothing coefficient, which prevents situations like zero denominators. The computation of <math display="inline">\mathsf{BCEDiceLoss}</math> is as follows:
+In Eq. (11), <math>\varepsilon</math> represents a smoothing coefficient, which prevents situations like zero denominators. The computation of <math display="inline">\mathsf{BCEDiceLoss}</math> is as follows:
 {| class="formulaSCP" style="width: 100%; text-align: left;"

@@ Line 257: / Line 257: @@
 |}
+In Eq. (11), <math>\epsilon</math> represents a smoothing coefficient, which prevents situations like zero denominators. The computation of <math display="inline">\mathsf{BCEDiceLoss}</math> is as follows:
-In Eq. (11), a represents a smoothing coefficient, which prevents situations like zero denominators. The computation of <math display="inline">\mathsf{BCEDiceLoss}</math> is as follows:
 {| class="formulaSCP" style="width: 100%; text-align: left;"

@@ Line 245: / Line 245: @@
 In Eq. (10), <math display="inline"> X </math> represents the initial thyroid nodule image, <math display="inline"> Y </math> the true labels, and <math display="inline">\hat{Y}</math> the corresponding predicted labels.
-The\mathsf{Dice} loss function is defined as:
+The <math>\mathsf{Dice}</math> loss function is defined as:
 {| class="formulaSCP" style="width: 100%; text-align: left;"

@@ Line 166: / Line 166: @@
 . Depth-wise separable convolution is advantageous for encoding positional information of features. Experimental results indicate that convolutional layers in MLPs are sufficient for encoding positional information and outperform standard positional encoding in practical performance.
-. DWConv has fewer parameters, In the tokenized MLP stage, features are initially transformed and projected onto tokens, with the channel count adjusted to match the number of tokens.
+. DWConv has fewer parameters. In the tokenized MLP stage, the features are initially transformed and projected onto tokens, with the channel count adjusted to match the number of tokens.
 The computational process of the tokenized MLP stage module involves:

@@ Line 23: / Line 23: @@
 Although thyroid nodule ultrasound imaging technology is mature, the quality of imaging cannot be guaranteed, and shortcomings such as blurred edges of thyroid nodules in images are unavoidable. Differences in the model and type of ultrasound equipment also lead to significant differences in the collected ultrasound images. Additionally, fine-needle aspiration biopsy surgery requires a large amount of medical and human resources and is somewhat invasive for patients. Therefore, this diagnostic method heavily relies on the subjective judgment of attending physicians, which can easily lead to misdiagnosis due to differences in doctors’ operational experience and techniques. Unnecessary biopsy surgeries can also cause patients more suffering.  Therefore, improving the accuracy of segmentation for ultrasound images of thyroid nodules in computational fields will notably enhance the precision and efficacy of clinical diagnosis and treatment.
-Addressing thyroid nodule segmentation, the U-Net model, as introduced by Ronneberger et al. [3], revolutionized deep learning techniques for medical image segmentation by integrating skip connections within its encoder-decoder architecture. This advancement marked a significant milestone, heralding a new era in the field. In a parallel development, Ma et al. [4] improved the network by introducing a multi-dilation convolutional block. This enhancement enables more accurate segmentation of nodule regions, resulting in the creation of more precise binary masks for medical image segmentation. Hu Yishan et al. [5] introduced attention mechanisms for thyroid nodule segmentation, optimizing low-dimensional features of images and preserving important features through the fusion of high and low-dimensional features. Sun et al. [6] fused different feature layers with the U-Net as the backbone network and introduced SE attention mechanisms to further improve segmentation accuracy. Chu et al. [7] introduced a thyroid nodule segmentation network utilizing U-Net architecture, substantially enhancing segmentation accuracy with limited datasets, thereby effectively aiding physicians in diagnosing thyroid nodules. Oktay et al. [8] introduced the Attention-UNet, a novel network model designed to automatically prioritize targets of diverse sizes and shapes. This approach effectively accentuates significant features while mitigating attention towards irrelevant areas. Zhou et al. [9] developed the Deeply Supervised Encoder-Decoder UNet++ network. This diminishes the semantic disparity between the feature maps of encoder and decoder subnetworks. Meanwhile, Chen et al. [10] enhanced the DeepLabv3+ model by integrating a decoder module to refine segmentation outcomes and integrating depth-wise separable convolutions into both the spatial pyramid pooling and decoder modules. Badrinarayanan et al. [11] introduced the SegNet segmentation network. It symmetrically performs downsampling and upsampling. Many models adopt multi-stage segmentation methods, further increasing computational complexity, indicating the need to improve the segmentation speed of many thyroid nodule models.
+Addressing thyroid nodule segmentation, the U-Net model, as introduced by Ronneberger et al. [3], revolutionized deep learning techniques for medical image segmentation by integrating skip connections within its encoder-decoder architecture. This advancement marked a significant milestone, heralding a new era in the field. In a parallel development, Ma et al. [4] improved the network by introducing a multi-dilation convolutional block. This enhancement enables more accurate segmentation of nodule regions, resulting in the creation of more precise binary masks for medical image segmentation. Hu  et al. [5] introduced attention mechanisms for thyroid nodule segmentation, optimizing low-dimensional features of images and preserving important features through the fusion of high and low-dimensional features. Sun et al. [6] fused different feature layers with the U-Net as the backbone network and introduced SE attention mechanisms to further improve segmentation accuracy. Chu et al. [7] introduced a thyroid nodule segmentation network utilizing U-Net architecture, substantially enhancing segmentation accuracy with limited datasets, thereby effectively aiding physicians in diagnosing thyroid nodules. Oktay et al. [8] introduced the Attention-UNet, a novel network model designed to automatically prioritize targets of diverse sizes and shapes. This approach effectively accentuates significant features while mitigating attention towards irrelevant areas. Zhou et al. [9] developed the Deeply Supervised Encoder-Decoder UNet++ network. This diminishes the semantic disparity between the feature maps of encoder and decoder subnetworks. Meanwhile, Chen et al. [10] enhanced the DeepLabv3+ model by integrating a decoder module to refine segmentation outcomes and integrating depth-wise separable convolutions into both the spatial pyramid pooling and decoder modules. Badrinarayanan et al. [11] introduced the SegNet segmentation network. It symmetrically performs downsampling and upsampling. Many models adopt multi-stage segmentation methods, further increasing computational complexity, indicating the need to improve the segmentation speed of many thyroid nodule models.
 Currently, the widely used deep learning neural network in medical imaging is the U-Net network. However, U-Net still faces limitations in thyroid nodule segmentation, such as ineffective utilization of pixel-space information and long training times. The primary contribution of this paper is the research and development of an optimal network structure designed to accurately segment nodules in the thyroid region.

@@ Line 455: / Line 455: @@
 . Combining the advantages of both <math>\mathsf{BCE}</math> and <math display="inline">\mathsf{Dice}</math> loss functions by using the <math>\mathsf{BCEDice}</math> loss function, which balances stability and segmentation accuracy, further enhancing the model’s performance.
-Experimental findings suggest that the enhanced network model proposed in this study attains superior segmentation accuracy metrics. Specifically, it achieves a <math display="inline">\mathsf{Dice}</math> coefficient of 0.9062, a precision rate of 0.9153, an average recall of 0.9023, and an average <math display="inline">F-1</math> score of 0.9.
+Experimental findings suggest that the enhanced network model proposed in this study attains superior segmentation accuracy metrics. Specifically, it achieves a <math display="inline">\mathsf{Dice}</math> coefficient of 0.9062, a precision rate of 0.9153, an average recall of 0.9023, and an average <math display="inline">F_1</math> score of 0.9.
 ==References==

@@ Line 458: / Line 458: @@
 ==References==
 [1] Guth S., Theune U., Aberle J., Galach A., Bamberger C.M. Very high prevalence of thyroid nodules detected by high frequency (13 MHz) ultrasound examination. Eur. J. Clin. Invest., 39:699–706, 2009.
-[2] Haugen BR, Alexander EK, Bible KC, et al. 2015 American Thyroid Association management guidelines for adult patients with thyroid nodules and differentiated thyroid cancer: the American Thyroid Association Guidelines Task Force on thyroid nodules and differentiated thyroid cancer. Thyroid 2016; 26: 01–133.
+[2] Haugen B.R., Alexander E.K., Bible K.C., et al. 2015 American Thyroid Association management guidelines for adult patients with thyroid nodules and differentiated thyroid cancer: the American Thyroid Association Guidelines Task Force on thyroid nodules and differentiated thyroid cancer. Thyroid, 26:1–133, 2016.
-[3] Ronneberger O, Fischer P, Brox T. U-net: Convolutional Networks for Biomedical Image Segmentation.  International Conference on Medical Image Computing and Computer-Assisted Intervention, Munich, Germany, 2015: 234-241.
+[3] Ronneberger O., Fischer P., Brox T. U-net: Convolutional networks for biomedical image segmentation.  International Conference on Medical Image Computing and Computer-Assisted Intervention, Munich, Germany, pp. 234-241, 2015.
-[4] Ma Xiaoxuan, Sun Boyang, Liu Weifeng, et al. AMSeg: A Novel Adversarial Architecture Based Multi-Scale Fusion Framework for Thyroid Nodule Segmentation. IEEE Access, 2023, 11: 72911-72924.
+[4] Ma X., Sun B., Liu W., et al. AMSeg: A novel adversarial architecture based multi-scale fusion framework for thyroid nodule segmentation. IEEE Access, 11:72911-72924, 2023.
-[5]  Hu Yishan, Qin Pinle, Zeng Jianchao, et al. Ultrasound thyroid segmentation network based on feature fusion and dynamic multi-scale dilated convolution. Journal of Computer Applications, 2021, 41(3): 891-897.
+[5]  Hu Y., Qin P., Zeng J., et al. Ultrasound thyroid segmentation network based on feature fusion and dynamic multi-scale dilated convolution. Journal of Computer Applications, 41(3):891-897, 2021.
-[6] Sun J, Li C, Lu Z, et al. TNSNet: Thyroid nodule segmentation in ultrasound imaging using soft shape supervision.  Computer Methods and Programs in Biomedicine, 2022, 215, 106600.
+[6] Sun J., Li C., Lu Z., et al. TNSNet: Thyroid nodule segmentation in ultrasound imaging using soft shape supervision.  Computer Methods and Programs in Biomedicine, 215, 106600, 2022.
-[7] Chu C, Zheng J, Zhou Y. Ultrasonic thyroid nodule detection method based on U-Net network. Computer Methods and Programs in Biomedicine, 2021, 199: 105906-105912.
+[7] Chu C., Zheng J., Zhou Y. Ultrasonic thyroid nodule detection method based on U-Net network. Computer Methods and Programs in Biomedicine, 199:105906-105912, 2021.
-[8] Oktay O, Schlemper J, Folgoc L L, et al. Attention u-net: Learning where to look for the pancreas. arXiv preprint arXiv:1804.03999, 2018.
+[8] Oktay O., Schlemper J., Folgoc L.L., et al. Attention u-net: Learning where to look for the pancreas. arXiv preprint arXiv:1804.03999, 2018.
-[9] Zhou Z, Rahman Siddiquee M M, Tajbakhsh N, et al. Unet++: A nested u-net architecture for medical image segmentation. Deep learning in medical image analysis and multimodal learning for clinical decision support. Springer, Cham, 2018: 3-11.
+[9] Zhou Z., Rahman Siddiquee M.M., Tajbakhsh N., et al. Unet++: A nested u-net architecture for medical image segmentation. In: Stoyanov, D., et al. Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support, DLMIA ML-CDS 2018 2018. Lecture Notes in Computer Science, vol. 11045, Springer, Cham., 3-11, 2018.
-[10] Chen L C, Zhu Y, Papandreou G, et al. Encoder-decoder with atrous separable convolution for semantic image segmentation. Proceedings of the European conference on computer vision (ECCV), Munich, Germany, 2018: 801-818.
+[10] Chen L.C., Zhu Y., Papandreou G., et al. Encoder-decoder with atrous separable convolution for semantic image segmentation. Proceedings of the European conference on computer vision (ECCV), Munich, Germany, 801-818, 2018.
-[11] Badrinarayanan V, Kendall A, Cipolla R. Segnet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39(12): 2481-2495.
+[11] Badrinarayanan V., Kendall A., Cipolla R. Segnet: A deep convolutional encoder-decoder architecture for image segmentation. IEEE Transactions on Pattern Analysis & Machine Intelligence, 39(12):2481-2495, 2017.
-[12] Dan Y, Jin W, Wang Z, et al. Optimization of U-shaped pure transformer medical image segmentation network. PEERJ Computer Science, 2023, 9, 1515.
+[12] Dan Y., Jin W., Wang Z., et al. Optimization of U-shaped pure transformer medical image segmentation network. PeerJ Computer Science, 9, 1515, 2023.

@@ Line 13: / Line 13: @@
 ==Abstract==
-To develop a precise neural network model designed for segmenting ultrasound images of thyroid nodules. The deep learning U-Net network model was utilized as the main backbone, with improvements made to the convolutional operations and the implementation of multilayer perceptron modeling at the lower levels, using the more effective <math display="inline">\mathsf{BCEDice}</math> loss function. The modified network achieved enhanced segmentation precision and robust generalization capabilities, with a <math>\mathtt{Dice}</math> coefficient of 0.9062, precision of 0.9153, recall of 0.9023, and an <math display="inline"> F_1 </math> score of 0.9062, indicating improvements over the U-Net and Swin-Unet to various extents. The U-Net network enhancement presented in this study outperforms the original U-Net across all performance indicators. This advancement could help physicians make more precise and efficient diagnoses, thereby minimizing medical errors.
+To develop a precise neural network model designed for segmenting ultrasound images of thyroid nodules. The deep learning U-Net network model was utilized as the main backbone, with improvements made to the convolutional operations and the implementation of multilayer perceptron modeling at the lower levels, using the more effective <math display="inline">\mathsf{BCEDice}</math> loss function. The modified network achieved enhanced segmentation precision and robust generalization capabilities, with a <math>\mathsf{Dice}</math> coefficient of 0.9062, precision of 0.9153, recall of 0.9023, and an <math display="inline"> F_1 </math> score of 0.9062, indicating improvements over the U-Net and Swin-Unet to various extents. The U-Net network enhancement presented in this study outperforms the original U-Net across all performance indicators. This advancement could help physicians make more precise and efficient diagnoses, thereby minimizing medical errors.
 '''Keywords''': U-Net, image segmentation, thyroid nodule ultrasound imaging, deep learning