Medical Diagonosis System Using AI Evolution Algorithms - CNN Based Chest X-ray Classification -
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
The application of artificial intelligence techniques, specifically evolution algorithms in medical diagnosis has shown promising potential to enhance the accuracy and efficiency of disease identification. These days the trend is evolutional algorithms be on the turn CNN((Conversion Neural Network) algorithms. As a result of that this paper research applied to medical devices using CNN technology is being actively carried out around the world. In this study, chest x-ray image classification is performed using CNN-based ResNet as a medical image reading assistance technology. In this study, we will classify pneumonia. A total of 5,863 x-ray images are used, and the data is divided into pneumonia and normal, and consists of 3,875 pneumonia image data and 1,341 normal image data. Shortcut is connected to a cnn composed of convolutional layers, pooling layers, and fully connected layers to improve performance using fine tuning on CNN-ResNet, which has been learned to minimize residual. In this paper, only the use of the pre-learning model was considered as fine tuning, but the batch size and learning rate also affect the learning of the model. It is expected that further research to find the appropriate proportions will allow the performance of the model to be maintained more stable.
Abstract
The application of artificial intelligence techniques, specifically evolution algorithms in medical diagnosis has shown promising potential to enhance the accuracy and efficiency of disease identification. These days the trend is evolutional algorithms be on the turn CNN((Conversion Neural Network) algorithms. As a result of that this paper research applied to medical devices using CNN technology is being actively carried out around the world. In this study, chest x-ray image classification is performed using CNN-based ResNet as a medical image reading assistance technology. In this study, we will classify pneumonia. A total of 5,863 x-ray images are used, and the data is divided into pneumonia and normal, and consists of 3,875 pneumonia image data and 1,341 normal image data. Shortcut is connected to a cnn composed of convolutional layers, pooling layers, and fully connected layers to improve performance using fine tuning on CNN-ResNet, which has been learned to minimize residual. In this paper, only the use of the pre-learning model was considered as fine tuning, but the batch size and learning rate also affect the learning of the model. It is expected that further research to find the appropriate proportions will allow the performance of the model to be maintained more stable.
참고문헌 (12)
- H. S. Lee, "The AI trend me dical trend based big data", KIICE, 37(9), pp.85~91 (2020)
- K. H. Jung, "The medical image analysis trend based AI", IITP Weekly trend, 22(8), pp.225~227 (2018)
- "Transfer Learning" (2023)[URL]
- "The Introduction of Natural Processing CNN" (2023)[URL]
- K. He, "Deep residual learning for image recognition", IEEE, 20(5), pp.125~137 (2016)
- D. K. Kim, "COVID-19 prediction using natural network compound product", Master’s, pp.2022
- 신형섭, "어텐션 기법 및 의료 영상에의 적용에 관한 최신 동향", 대한영상의학회지, 81(6), pp.1305~1333 (2020), 대한영상의학회[DOI]
- U-Net convolutional networks for biomedical image segmentation, "O. Ronneberger", MICCAI, 2852~2860 (2015)
- A. Vaswani, "Attention is all you need", 8th conference, pp.152~159 (2017)
- Attention U-Net:learning where to look for the pancreas, "O. Oktay", MICCAI, 219~228 (2018)
- F. Wang, "Residual attention network for image classification", Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition(CVPR) (2017)
- J. Schlemper, "Attention gated network learning to leverage salient regions in medi ical images", 53, pp.197~207 (2019)