http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
선명도에 따른 조직 이미지 분류: 딥러닝 모델과 이미지 품질 측정 기법의 성능 비교
송진솔(Jin Sol Song),변근호(Keun Ho Byeon),임지우(Jee Woo Lim),이재웅(Jae Ung Lee),박선홍(Sun Hong Park),채승완(Chae Seoung Wan),곽진태(Jin Tae Kwak) 대한전자공학회 2024 대한전자공학회 학술대회 Vol.2024 No.6
In digital pathology, most of artificial intelligence techniques are built based upon high-quality, sharp tissue images. However, digitized tissue images often contain blurry regions that can hinder accurate and precise analysis and decision making on the tissue images. In this paper, we develop deep learning models that can distinguish sharp tissue images from blurry tissue images. Five different deep learning models are employed and compared to five conventional image quality assessment methods using >245,000 tissue image patches. The experimental results demonstrate that deep learning models outperform image quality assessment techniques with substantial reduction in processing time. The best deep learning model achieves 99.57% accuracy and 0.9973 F1-score. These suggest that the deep learning models are promising for assessing sharpness in tissue images.
의료 영상 병변 영상 분할을 위한 효율적 딥러닝 모델 연구
심하민(Hamin Shim),양현(Hyun Yang),정수민(Sumin Jang),곽진태(Jin Tae Kwak) 대한전자공학회 2024 대한전자공학회 학술대회 Vol.2024 No.6
In this study, we explored deep learning models for automated lesion segmentation in medical imaging. Encoder-decoder frameworks are commonly employed for image segmentation. Herin, we utilized two CNN-based models (3D UNet and UNet++) and one transformer-based model (3D MobileViTv3). These encoder-decoder neural networks were applied to two types of medical image segmentation: brain tumor segmentation observed in MRI scans and liver tumor segmentation observed in CT scans. Our experimental findings demonstrate that while deep learning models can successfully segment tumor regions, their performance varies based on model structures.
다중 기관에서의 디지털 병리 암 분화도 예측을 위한 멀티 태스크 기반 단일 모델 학습
임종우(Lim Jong Woo),신상혁(Shin Sang Hyeok),강동연(Kang Dong Yeon),이주천(Jucheon Lee),이재웅(Jaeung Lee),곽진태(Jin Tae Kwak) 대한전자공학회 2023 대한전자공학회 학술대회 Vol.2023 No.11
In this study, we propose a single multi-task deep learning model for classifying digital pathology images from multiple organs based on the degree of cancer differentiation. For multi-organ cancer classification, there has been two major approaches in digital pathology. One is to develop a separate model per organ. Second is to employ an ensemble model to combine multiple models that were trained on different organs. Both approaches are time- and resource-inefficient. Herein, we propose a single multi-task model that simultaneously utilizes pathology images from multiple organs. Three digital pathology datasets, including colon, prostate, and gastric tissue images, are employed in this study. The experimental results demonstrate that the proposed approach is able to improve the overall cancer classification performance, which outperforms single organ models and ensemble models.
디지털 병리 대장암 진단을 위한 삼중항 손실을 이용한 심층 메트릭 학습
이재웅(Jae Ung Lee),김경은(Kyung Eun Kim),송보람(Bo Ram Song),이주천(Ju Cheon Lee),Vuong Thi Le Trinh,Wang Jiamu,Syed Farhan Abbas,곽진태(Jin Tae Kwak) 대한전자공학회 2022 대한전자공학회 학술대회 Vol.2022 No.11
In this paper, we propose a deep metric learning classification model for colon cancer grading which aims to learn feature vector similarity by distance comparisons. The proposed method is evaluated on >9,800 colorectal image patches. The experimental results show that the method achieves 87.85% accuracy and 0.8425 F1-score, suggesting that the proposed learning method can improve histopathological analysis of cancer grade classification in pathological images.