scispace - formally typeset
Search or ask a question
Author

Junying Chen

Bio: Junying Chen is an academic researcher from South China University of Technology. The author has contributed to research in topics: Deep learning & Active contour model. The author has an hindex of 1, co-authored 3 publications receiving 3 citations.

Papers
More filters
Proceedings ArticleDOI
01 Oct 2017
TL;DR: The segmentation results shown in the paper demonstrated that the combined image segmentation algorithms and image preprocessing methods can successfully segment the thyroid regions out of thyroid ultrasound images.
Abstract: Image segmentation for thyroid ultrasound images is a challenging task. As such, several image segmentation algorithms combined with different image preprocessing methods applied to thyroid ultrasound image segmentation are studied in this work. The image segmentation algorithms presented in this paper include edge detection, regional segmentation and active contour without edge algorithms. The image preprocessing methods presented in this paper include Butterworth low-pass filtering, Butterworth high-pass enhanced filtering, and adaptive weighted median filtering. In the experiments, the image segmentation algorithms and image preprocessing methods were combined to evaluate the segmentation results for thyroid ultrasound images. The segmentation results shown in the paper demonstrated that the combined image segmentation algorithms and image preprocessing methods can successfully segment the thyroid regions out of thyroid ultrasound images.

3 citations

Proceedings ArticleDOI
11 Sep 2021
TL;DR: In this paper, an auto-encoding generative adversarial network combining the advantages of GAN and VAE was used to generate realistic ultrasound features and thyroid tissues for augmentation and help training a U-Net model to get better segmentation results.
Abstract: Ultrasound (US) has been investigated as a common method of computer aided diagnosis because of its low-cost, harmless and real-time scanning. Also the rapid development of deep learning segmentation and classification models alleviates the influence of low signal-to-noise ratio and artifacts of ultrasonic imaging. However, due to the privacy issues of medical data, it is not easy to acquire sufficient data for deep learning model training. In recent years, generative adversarial networks (GANs) are widely used in data augmentation. However, GANs suffer from the problem of mode collapse in the training process then generate images with a limited variety. On the other hand, variational auto-encoder (VAE) is free from mode collapse but it generates blurred images. In this work, we study an auto-encoding generative adversarial network combining the advantages of GAN and VAE to generate realistic images for medical thyroid ultrasound image augmentation. Experiment results show that the generated images can simulate realistic ultrasound features and thyroid tissues for augmentation and help training a U-Net model to get better segmentation results.

3 citations

Proceedings ArticleDOI
11 Sep 2021
TL;DR: In this paper, the relative position relationship between different tissues in ultrasound images is added to the loss function as a prior knowledge to improve the semantic expression of loss function in measuring the difference between the predicted results and the ground-truth labels.
Abstract: In deep learning, loss function plays a crucial role in training an effective neural network model. In the task of ultrasound image segmentation, the pixel-wise loss functions such as cross-entropy loss and dice loss are usually used to train a deep neural network model. These loss functions only count the distribution differences between the predicted results of the model and the ground-truth labels at pixel level, but do not pay attention to the consistence of the spatial relations between different tissues in the predicted results and the real images. In order to improve the semantic expression of the loss function in measuring the difference between the predicted results and the ground-truth labels, we use the concept of relative fuzzy connectedness to add the relative position relationship between tissues in ultrasound images to the loss function as a prior knowledge.

Cited by
More filters
Journal ArticleDOI
TL;DR: A comprehensive and systematic review and analysis of medical image augmentation work are carried out, and its research status and development prospects are reviewed in this paper , which mainly collected by ELSEVIER, IEEE Xplore, and Springer from 2018 to 2021.

98 citations

Journal ArticleDOI
TL;DR: A comprehensive review of the use of deep generative models for medical image augmentation can be found in this paper , where the authors highlight the potential of these models for improving the performance of deep learning algorithms in medical image analysis.
Abstract: Deep learning has become a popular tool for medical image analysis, but the limited availability of training data remains a major challenge, particularly in the medical field where data acquisition can be costly and subject to privacy regulations. Data augmentation techniques offer a solution by artificially increasing the number of training samples, but these techniques often produce limited and unconvincing results. To address this issue, a growing number of studies have proposed the use of deep generative models to generate more realistic and diverse data that conform to the true distribution of the data. In this review, we focus on three types of deep generative models for medical image augmentation: variational autoencoders, generative adversarial networks, and diffusion models. We provide an overview of the current state of the art in each of these models and discuss their potential for use in different downstream tasks in medical imaging, including classification, segmentation, and cross-modal translation. We also evaluate the strengths and limitations of each model and suggest directions for future research in this field. Our goal is to provide a comprehensive review about the use of deep generative models for medical image augmentation and to highlight the potential of these models for improving the performance of deep learning algorithms in medical image analysis.

4 citations

Journal ArticleDOI
TL;DR: In this paper, a deep learning-based framework for diagnosing human malaria infection from microscopic images of thin blood smears is presented, which is based on a direct segmentation and classification approach which relies on the analysis of the parasite itself.
Abstract: Malaria is an infectious disease caused by Plasmodium parasites and is potentially human life-threatening. Children under 5 years old are the most vulnerable group with approximately one death every two minutes, accounting for more than 65% of all malaria deaths. The World Health Organization (WHO) encourages the research of appropriate methods to treat malaria through rapid and economical diagnostic. In this paper, we present a deep learning-based framework for diagnosing human malaria infection from microscopic images of thin blood smears. The framework is based on a direct segmentation and classification approach which relies on the analysis of the parasite itself. The framework permits to segment the Plasmodium parasite in the images and to predict its species among four dominant classes: P. Falciparum, P. Malaria, P. Ovale, and P. Vivax. A high potential of generalization with a competitive performance of our framework on inter-class data is demonstrated through an experimental study considering several datasets. Our source code is publicly available on https://github.com/Benhabiles-JUNIA/MalariaNet.

2 citations

04 Feb 2020
TL;DR: By implementing this system, the time for the diagnosis of malaria will be cut down, which will save lives and the medical resources that are used while waiting for the results of the tests from the old system.
Abstract: As technology has evolved it has become more and more efficient to diagnose, and treat multiple diseases. Malaria is one of the deadliest diseases on this pl anet. Each year it estimated that 1 million people die as a re sult of this disease. Furthermore 3.4 billion people are in dang er of contracting malaria. With advances in the field of medicine it is now entirely possible to not only tr eat but also prevent malaria. The way in which people are diagnosed for malaria today is through blood sample s. The techniques used currently are accurate however they are time consuming. This has necessitated doctors to st art the treatment for malaria before the blood work is finished, since in its later stage’s malaria can be very diff icult to cure. The system that is discussed aims to cut this time requirement by at least half and increase the accur acy of the tests. An automated system that gathers the ima ge data and analyses the images for malarial parasites is described. A system for collection of data and anal ysis is described. By implementing this system, the time ne eded for the diagnosis of malaria will be cut down. This will save lives and the medical resources that are used while waiting for the results of the tests from the old system. Keywords—Deep Learning, Malaria, Parasite

1 citations

Book ChapterDOI
16 May 2018
TL;DR: In this paper, the performance of image enhancement techniques on ultrasound images are evaluated using quality metrics, namely Mean Square Error (MSE) and Peak Signal-to-Noise Ratio (PSNR).
Abstract: Ultrasound imaging has been widely used in the diagnosis of atherosclerosis. To precisely diagnose the carotid plaque, the affected region should be segmented from the ultrasonic image of carotid artery. Many techniques have been used to identify the plaque in ultrasound images. Image enhancement and restoration are the important processes to acquire high-quality images from the noisy images. When the artery images are captured, noise occurs due to high-frequency rate. To acquire a high-quality image, preprocessing is the first step to be done. The quality of the image is improved in this process. The techniques involved in preprocessing are dealt in this paper. Preprocessing involves filtering the image and removing the noise by various filtering techniques. Salt-and-pepper and Gaussian noise in ultrasound images can be filtered using techniques like mean, median and Wiener filters. Salt-and-pepper noise is multiplicative in nature and it is introduced by the image acquisition mechanism. The quality of the input sensor is reflected by the Gaussian noise. In this paper, the performance of image enhancement techniques on ultrasound images are evaluated using quality metrics, namely Mean Square Error (MSE) and Peak Signal–to-Noise Ratio (PSNR).

1 citations