Wrist Ultrasound Segmentation by Deep Learning

Abstract

Supervised deep learning offers great promise to automate analysis of medical images from segmentation to diagnosis. However, their performance highly relies on the quality and quantity of the data annotation. Meanwhile, curating large annotated datasets for medical images requires a high level of expertise, which is time-consuming and expensive. Recently, to quench the thirst for large data sets with high-quality annotation, self-supervised learning (SSL) methods using unlabeled domain-specific data, have attracted attention. Therefore, designing an SSL method that relies on minimal quantities of labeled data has far-reaching significance in medical images. This paper investigates the feasibility of deploying the Masked Autoencoder for SSL (SSL-MAE) of TransUNet, for segmenting bony regions from children's wrist ultrasound scans. We found that changing the embedding and loss function in SSL-MAE can produce better downstream results compared to the original SSL-MAE. In addition, we determined that only pretraining TransUNet embedding and encoder with SSL-MAE does not work as well as TransUNet without SSL-MAE pretraining on downstream segmentation tasks.

Generative Adversarial Network (Generative Artificial Intelligence) in Pediatric Radiology: A Systematic Review

“…All identified articles were published over the last five years and the publication number increased every year with the highest number in 2022 [ 77 , 78 , 79 , 80 , 81 , 82 , 83 , 84 , 85 , 86 , 87 , 88 , 89 , 90 , 91 , 92 , 93 , 94 , 95 , 96 , 97 , 98 , 99 , 100 , 101 , 102 , 103 , 104 , 105 , 106 , 107 , 108 , 109 , 110 , 111 , 112 , 113 ]. This increasing trend was in line with the one in radiology [ 1 , 77 , 78 , 79 , 80 , 81 , 82 , 83 , 84 , 85 , 86 , 87 , 88 , 89 , 90 , 91 , 92 ,…”

“…All low-quality ones were conference papers ( n = 12) [ 79 , 80 , 81 , 83 , 85 , 86 , 91 , 93 , 94 , 95 , 104 , 108 ]. The GAN was commonly applied to MRI ( n = 18) [ 77 , 78 , 83 , 84 , 87 , 90 , 97 , 101 , 103 , 104 , 105 , 106 , 108 , 109 , 110 , 111 , 112 , 113 ] and X-ray ( n = 13) [ 79 , 80 , 89 , 91 , 92 , 94 , 95 , 96 , 98 , 99 , 100 , 102 , 107 ], and the others included computed tomography (CT) ( n = 4) [ 82 , 86 , 93 , 97 ], ultrasound ( n = 2) [ 85 , 88 ] and positron emission tomography (PET) ( n = 1) [ 81 ]. Although the basic GAN architecture was still popular among the included studies ( n = 11) [ 77 …”

“…Both image synthesis and data augmentation ( n = 12) [ 89 , 90 , 91 , 92 , 93 , 94 , 95 , 96 , 97 , 98 , 99 , 100 ], and image translation ( n = 12) [ 101 , 102 , 103 , 104 , 105 , 106 , 107 , 108 , 109 , 110 , 111 , 112 ] were the commonest application areas of GAN in pediatric radiology. Other GAN application areas included image segmentation ( n = 5) [ 84 , 85 , 86 , 87 , 88 ], image reconstruction ( n = 4) [ 80 , 81 , 82 , 83 ], disease diagnosis ( n = 3) [ 77 , 78 , 79 ], and image quality assessment ( n = 1) [ 113 ]. However, none of the GAN models involved in these studies were commercially available [ 77 , 78 , 79 , 80 , 81 , 82 , 83 , 84 , 85 , 86 , 87 , 88 , 89 ,…”

“…Other GAN application areas included image segmentation ( n = 5) [ 84 , 85 , 86 , 87 , 88 ], image reconstruction ( n = 4) [ 80 , 81 , 82 , 83 ], disease diagnosis ( n = 3) [ 77 , 78 , 79 ], and image quality assessment ( n = 1) [ 113 ]. However, none of the GAN models involved in these studies were commercially available [ 77 , 78 , 79 , 80 , 81 , 82 , 83 , 84 , 85 , 86 , 87 , 88 , 89 , 90 , 91 , 92 , 93 , 94 , 95 , 96 , 97 , 98 , 99 , 100 , 101 , 102 , 103 , 104 , 105 , 106 , 107 , 108 , 109 , 110 , 111 , 112 , 113 ]. For the twenty-nine studies which compared their GAN model performances with those of other approaches, all of them outperformed the others by 0.1–158.6% [ …”

Knee Bone Models From Ultrasound

“…Further approaches include the use of conditional generative adversarial networks (cGANs) like the pix2pix architecture introduced by Isola et al [59] for image-to-image translation. Zhou et al [60] employ pix2pix for direct segmentation and for the improvement of U-Net segmentations. Similarly, Alsinan et al [61] condition a cGAN for the generation of bone shadow maps, which are then fused with CNN-based features.…”