Figure - available from: Signal Image and Video Processing
This content is subject to copyright. Terms and conditions apply.
The process of acquiring the intersection point and the α angle in an ultrasound image. a Raw image, b result after performing local mean filtering using 10 × 1 window, c local contrast enhancement result, d the hip joint baseline after binarization and removal of smaller connected areas. Then, using a bottom-to-top search strategy to locate the ilium bone area, e contrast enhancement to the area below the ilium, f resetting the threshold value and binarizing the area below the ilium, g selecting a sufficient number of points in the ilium and acetabular bone area located in (d) and (e) and fitting them by least squares operation and h the fitting results that can be used to calculate the intersection point
Source publication
Developmental dysplasia of the hip is a medical term representing the hip joint instability that appears mainly in infants. The assessment metric of physician is based on the femoral head coverage rate, which needs to segment the femoral head area in 2D ultrasound images. In this paper, we propose an approach to automatically segment the femoral he...
Citations
... Models (e.g., fully convolutional networks (FCN) [7], U-Net [8], SegNet [9], and DeepLab [10][11][12][13]) have demonstrated their effectiveness in accurately segmenting images. Applications derived from deep-learning-based image segmentation have thrived in various fields, e.g., clinical medical image segmentation [14,15] and eye image segmentation [16]. ...
Accurate microscopic images segmentation of activated sludge is essential for monitoring wastewater treatment processes. However, it is a challenging task due to poor contrast, artifacts, morphological similarities, and distribution imbalance. A novel image segmentation model (FafFormer) was developed in the work based on Transformer that incorporated pyramid pooling and flow alignment fusion. Pyramid Pooling Module was used to extract multi-scale features of flocs and filamentous bacteria with different morphology in the encoder. Multi-scale features were fused by flow alignment fusion module in the decoder. The module used generated semantic flow as auxiliary information to restore boundary details and facilitate fine-grained upsampling. The Focal–Lovász Loss was designed to handle class imbalance for filamentous bacteria and flocs. Image-segmentation experiments were conducted on an activated sludge dataset from a municipal wastewater treatment plant. FafFormer showed relative superiority in accuracy and reliability, especially for filamentous bacteria compared to existing models.
... The only study using MRI data from patients younger than 16 years of age achieved the lowest DSC of 0.90, highlighting the potential difficulty of segmenting images from a younger age group [14]. Using ultrasound images of an unspecified age group, Chen et al. [15] segmented the femoral head with a DSC of 0.946. Finally, Jodeiri et al. [16] achieved a DSC of 0.96 when segmenting the entire pelvis from radiographs of patients requiring hip replacements. ...
... Compared with the segmentation method using CNNs, FCNs have two strong points: the acceptance of input images with any size and more efficient computation, which are widely used for ultrasound image segmentation. Chen et al. [34] propose an improved FCN named Fnet to automatically segment the femoral head in two-dimensional ultrasound images and achieved highaverage Dice, Recall and IoU scores. Sun et al. [35] develop FCN-AlexNet model structure and utilize transfer learning to realize the better performance of thyroid ultrasound image location and the diagnosis of benign or malignant lesions. ...
... Wijata et al. [32] suggest using a CNN to segment breast cancer ultrasonic images automatically. Chen et al. [34] propose an improved FCN to segment the bones in ultrasound images automatically. In the experiment of this paper, we reapply the two literature methods to compare with our proposed method on the self-built dataset. ...
... Dice comparison Firstly, we analyze the Dice evaluation of the experimental results, shown in the left part of [34]. Figure 5a describes the Boxplots of Dice score. ...
Image segmentation plays a critical role in the quantitative and qualitative analysis of medical ultrasound images, directly affecting the follow-up analysis and processing. However, due to the speckle noise, fuzziness, complexity and diversity of medical ultrasound images, the traditional image segmentation algorithms are accessible to leak the boundary at the weak edge of the medical ultrasound image, getting inaccurate results and difficulty extracting the target contour of the ultrasound image. In addition, the non-automatic feature extraction method cannot realize the end-to-end automatic segmentation function. Nevertheless, fully convolutional networks (FCNs) can realize end-to-end automatic semantic segmentation, and are widely used for ultrasound image segmentation. In this paper, we aim at the problems of low segmentation accuracy and long segmentation time in the traditional segmentation method, proposing a novel segmentation method based on an improved FCN with multi-scale dilated convolution for ultrasound image segmentation. The proposed method firstly preprocesses medical ultrasound images through image filtering, normalization and enhancement, and then improves the fully convolutional neural network by constructing four-dilated convolutions with different dilation rates, which can capture multi-scale context feature information and finally postprocesses the segmentation results of the medical ultrasound image by the Laplace correction operator. Our experiments demonstrate that the proposed method achieves better segmentation results than state-of-the-art methods on the breast ultrasound dataset.
... In the first method, mean filtering, morphological processing and least squares operation were used while in the second method, a CNN named FNet was utilized. In conclusion, the proposed deep neural network architecture method was found to provide better segmentation than other methods (Chen et al., 2020). ...
The term developmental dysplasia of the hip (DDH) describes a range of hip abnormalities affecting newborns where the femoral head and acetabulum are in improper alignment or grow abnormally, or both. The ultrasonographic evaluation technique rely on the capability of the ultrasonographer to pick up the accurate frame used for exact calculations. In our study we developed a new computer aided system that determines the exact frame from real time 2D ultrasound images and calculates the accuracy rate for each result. The deep learning architectures recently used in literature were utilized for these processes. In addition, transfer learning was carried out to increase the performance of the system using pretrained networks (SqueezeNet, VGG16, VGG19, ResNet50 and ResNet101). One of the best methods of object detection, You Only Look Once (YOLO) model, was used with pre-trained networks to determine DDH location. As a result of the study, the performance of the deep neural network model proposed with the help of these pre-trained networks was evaluated. When the obtained results were compared with expert opinions, frames (standard planes) in 605 of 676 (89.05%) test images were correctly detected. The accuracy rates for the used pre-trained networks were obtained as SqueezeNet 0.79, VGG16 0.95, VGG19 0.96, ResNet50 0.88 and ResNet101 0.93.
Ultrasound (US) is generally preferred because it is of low-cost, safe, and non-invasive. US image segmentation is crucial in image analysis. Recently, deep learning-based methods are increasingly being used to segment US images. This survey systematically summarizes and highlights crucial aspects of the deep learning techniques developed in the last five years for US segmentation of various body regions. We investigate and analyze the most popular loss functions and metrics for training and evaluating the neural network for US segmentation. Furthermore, we study the patterns in neural network architectures proposed for the segmentation of various regions of interest. We present neural network modules and priors that address the anatomical challenges associated with different body organs in US images. We have found that variants of U-Net that have dedicated modules to overcome the low-contrast and blurry nature of images are suitable for US image segmentation. Finally, we also discuss the advantages and challenges associated with deep learning methods in the context of US image segmentation.
