Ma Yinghua
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Ma Yinghua has a total of 10 co-authors. Here are the top co-authors by number of shared publications:
- Fiaz Gul Khan
Department of Computer Science, COMSATS University Islamabad, Abbottabad, Pakistan
2 shared publications - Yang Heng
Affiliation Department of Computer Science, COMSATS University Islamabad Abbottabad Campus, Abbottabad, Pakistan
2 shared publications Zeng Hui
Department of Computer Science, COMSATS University Islamabad, Abbottabad Campus, Abbottabad, Pakistan
2 shared publications- Ahmad Ali AlZubi
King Saud University
1 shared publication Ahmad Khan
Department of Computer Science, COMSATS University Islamabad, Abbottabad Campus, Abbottabad, Pakistan
1 shared publication- Ahmad Khan
Affiliation Department of Computer Science, COMSATS University Islamabad Abbottabad Campus, Abbottabad, Pakistan
1 shared publication
Latest Publications
- Open Access
- 20 December 2024
- Published by Springer Nature in Artificial Intelligence Review
- Vol. 58 (2) , 1-81
Survey: application and analysis of generative adversarial networks in medical images
- 20 December 2024
- Published by Springer Nature in Artificial Intelligence Review
- Vol. 58 (2) , 1-81
Open AccessSurvey: application and analysis of generative adversarial networks in medical images
- Open Access
- 6 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 126486-126502
Hard Exudates Segmentation in Diabetic Retinopathy Using DiaRetDB1
Diabetic retinopathy (DR) poses a major challenge in vision care, often leading to partial or complete blindness in adults. Early and accurate detection of DR is essential for timely intervention and effective patient management. One critical aspect of DR diagnosis is identifying hard exudates in retinal images. Using deep tech or computer-aided methods, ophthalmologists can gain detailed insights into retinal health, facilitating precise diagnosis and treatment planning. However, the availability of medical image data in open datasets is often limited, making deep analysis challenging. To address this issue, we introduce a UNet deep learning model, specifically modified and designed to segment hard exudates effectively, even with a limited dataset. In our study, the model is trained and tested on 89 original images from the DiaRetDB1 dataset, supplemented with augmented image to improve robustness. Despite the small dataset, our modified U-Net achieves high sensitivity, specificity, precision, and accuracy, culminating in an impressive F1 score of 0.9754. These results demonstrate the potential of our approach to improve DR diagnosis with limited resources. By adapting existing deep learning models, our research shows significant enhancements in diagnostic accuracy for DR. However, further validation with larger datasets and comparisons with other methods are necessary to confirm our findings. This study contributes to better detection of hard exudates in DR, ultimately leading to improved patient outcomes.
- 6 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 126486-126502
Open AccessHard Exudates Segmentation in Diabetic Retinopathy Using DiaRetDB1
Diabetic retinopathy (DR) poses a major challenge in vision care, often leading to partial or complete blindness in adults. Early and accurate detection of DR is essential for timely intervention and effective patient management. One critical aspect of DR diagnosis is identifying hard exudates in retinal images. Using deep tech or computer-aided methods, ophthalmologists can gain detailed insights into retinal health, facilitating precise diagnosis and treatment planning. However, the availability of medical image data in open datasets is often limited, making deep analysis challenging. To address this issue, we introduce a UNet deep learning model, specifically modified and designed to segment hard exudates effectively, even with a limited dataset. In our study, the model is trained and tested on 89 original images from the DiaRetDB1 dataset, supplemented with augmented image to improve robustness. Despite the small dataset, our modified U-Net achieves high sensitivity, specificity, precision, and accuracy, culminating in an impressive F1 score of 0.9754. These results demonstrate the potential of our approach to improve DR diagnosis with limited resources. By adapting existing deep learning models, our research shows significant enhancements in diagnostic accuracy for DR. However, further validation with larger datasets and comparisons with other methods are necessary to confirm our findings. This study contributes to better detection of hard exudates in DR, ultimately leading to improved patient outcomes.
- Open Access
- 17 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 55448-55464
HLSNC-GAN: Medical Image Synthesis Using Hinge Loss and Switchable Normalization in CycleGAN
In the field of medical image analysis, MRI and CT, among other multimodal medical images, play crucial roles. To overcome the limitations of image acquisition, researchers have proposed medical image synthesis techniques, including both traditional methods and deep learning approaches. In this study, we introduce a universal framework based on cycleGAN for generating CT images from MRI data.This framework incorporates a hinge loss function to establish mappings between different modalities and enhance structural consistency between input and output images. We also employ a switchable normalization technique to improve model stability and reduce manual intervention. These enhancements result in the generation of higher-quality synthetic images while avoiding gradient issues and mode collapse.The results of this research demonstrate significant progress in medical image synthesis. Compared to existing methods, our model exhibits superior performance in quantitative evaluation metrics while maintaining better diversity and structural consistency. This indicates that our framework holds promise in medical image synthesis and can provide valuable support in areas such as disease prediction and treatment.
- 17 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 55448-55464
Open AccessHLSNC-GAN: Medical Image Synthesis Using Hinge Loss and Switchable Normalization in CycleGAN
In the field of medical image analysis, MRI and CT, among other multimodal medical images, play crucial roles. To overcome the limitations of image acquisition, researchers have proposed medical image synthesis techniques, including both traditional methods and deep learning approaches. In this study, we introduce a universal framework based on cycleGAN for generating CT images from MRI data.This framework incorporates a hinge loss function to establish mappings between different modalities and enhance structural consistency between input and output images. We also employ a switchable normalization technique to improve model stability and reduce manual intervention. These enhancements result in the generation of higher-quality synthetic images while avoiding gradient issues and mode collapse.The results of this research demonstrate significant progress in medical image synthesis. Compared to existing methods, our model exhibits superior performance in quantitative evaluation metrics while maintaining better diversity and structural consistency. This indicates that our framework holds promise in medical image synthesis and can provide valuable support in areas such as disease prediction and treatment.