Segmentation of Lymph Nodes in Ultrasound Images Using U-Net Convolutional Neural Networks and Gabor-Based Anisotropic Diffusion

Abstract

Objective Automated segmentation of lymph nodes (LNs) in ultrasound images is a challenging task mainly due to the presence of speckle noise and echogenic hila. In this paper, we propose a fully automatic and accurate method for LN segmentation in ultrasound. MethodsThe proposed segmentation method integrates diffusion-based despeckling, U-Net convolutional neural networks and morphological operations. Firstly, we suppress speckle noise and enhance lymph node edges using the Gabor-based anisotropic diffusion (GAD). Secondly, a modified U-Net model is proposed to segment LNs excluding echogenic hila. Finally, morphological operations are adopted to segment entire LNs by filling the regions of echogenic hila.Results A total of 531 lymph nodes from 526 patients were included to evaluate the proposed method. Quantitative metrics of segmentation performance, including the accuracy, sensitivity, specificity, Jaccard similarity and Dice coefficient, reached 0.934, 0.939, 0.937, 0.763 and 0.865, respectively. ConclusionThe proposed method automatically and accurately segments LNs in ultrasound, which may assist artificially intelligent diagnosis of lymph node diseases.

CADNet: an advanced architecture for automatic detection of coronary artery calcification and shadow border in intravascular ultrasound (IVUS) images

“…The U-Net architecture developed, is the first model for biomedical image segmentation. Most of the medical imaging tasks apply encoder-decoder based architectures for the segmentation of complex structures [ 27 , 42 , 51 , 66 , 71 , 72 ]. For detecting the vessel components in IVUS images, we prefer to enhance the capability of the encoder-decoder based U-Net architecture with additions of a lightweight attention mechanism (to give more focus to crucial areas) and dilated convolutions (to extract multi-scale features with varying dilation rates of {1, 6, 12, 18}).…”

“…Due to the scarcity of a limited amount of annotated data in the case of biomedical imaging, data augmentation is preferred [ 61 ]. Various augmentation (operation) types have been investigated in the literature to create fresh training images: When analysing visual documents, elastic distortions are used [ 62 ]; other examples include scaling, translation, shearing, flipping, and rotation transformations [ 19 , 20 , 27 , 39 , 51 , 63 – 66 ]. However, not every augmentation operation will be useful in medical settings.…”

Performance analysis of various denoising filters on intravascular ultrasound coronary artery images

“…NLM filter with improvements is applied in recent years for ultrasound images which resulted in better outputs 27,28 . Recently, a Gabor‐Based Anisotropic Diffusion filter was used for speckle noise suppression in lymph node segmentation before training the U‐Net model, and the results were better than the model, trained using raw speckled corrupted images 29 . A few years back, China et al, 1 proposed an algorithm using wavelets and the NLM filter, utilizing the approximation coefficients, which underwent the filtering process while preserving the detailed coefficients for IVUS images.…”

“…27,28 Recently, a Gabor-Based Anisotropic Diffusion filter was used for speckle noise suppression in lymph node segmentation before training the U-Net model, and the results were better than the model, trained using raw speckled corrupted images. 29 A few years back, China et al, 1 proposed an algorithm using wavelets and the NLM filter, utilizing the approximation coefficients, which underwent the filtering process while preserving the detailed coefficients for IVUS images. Despeckling in IVUS images was further improved using curvelet transforms and adaptive complex diffusion filters.…”