A. Kakumani, L. P. Sree, Chittimalla Sai Krishna, Greeshmitha Uppalapati, Gudimetla Santhoshi Sri Pavithra, Seshadri Harshini
{"title":"Semantic Segmentation of Cells in Microscopy Images via Pretrained Autoencoder and Attention U-Net","authors":"A. Kakumani, L. P. Sree, Chittimalla Sai Krishna, Greeshmitha Uppalapati, Gudimetla Santhoshi Sri Pavithra, Seshadri Harshini","doi":"10.1109/MLCSS57186.2022.00025","DOIUrl":null,"url":null,"abstract":"Cell segmentation is one of the most important steps for medical image examination. With the rapid advancement of the convolutional neural networks, deep learning-based approach has become popular for the segmentation and analyzing the cells. Despite this, accurate, general, and consistent segmentation is still required to accurately evaluate its morphological features, phenotypes, and sub-cellular dynamics. Previously U-Net architecture approaches have been proposed. However, a loss of some spatial information is observed. To enhance the performance of U-Net on various segmentation challenges, we have come up with a novel architecture called Pretrained Autoencoder with Attention U-N et. Trained autoencoder weights are assigned to the contraction pathway of U-Net and in addition to that attention block is added to the expansion pathway of the vanilla U-Net. The auto-encoder helps in pretraining the weights of the pixels and the attention block pays more attention to the relevant regions. Our approach achieved 98.27% accuracy when it came to segmenting the cells, it outperformed the state-of-the-art U-N et in terms of traditional parameters.","PeriodicalId":322975,"journal":{"name":"2022 International Conference on Machine Learning, Computer Systems and Security (MLCSS)","volume":"34 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2022-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"1","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2022 International Conference on Machine Learning, Computer Systems and Security (MLCSS)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/MLCSS57186.2022.00025","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 1
Abstract
Cell segmentation is one of the most important steps for medical image examination. With the rapid advancement of the convolutional neural networks, deep learning-based approach has become popular for the segmentation and analyzing the cells. Despite this, accurate, general, and consistent segmentation is still required to accurately evaluate its morphological features, phenotypes, and sub-cellular dynamics. Previously U-Net architecture approaches have been proposed. However, a loss of some spatial information is observed. To enhance the performance of U-Net on various segmentation challenges, we have come up with a novel architecture called Pretrained Autoencoder with Attention U-N et. Trained autoencoder weights are assigned to the contraction pathway of U-Net and in addition to that attention block is added to the expansion pathway of the vanilla U-Net. The auto-encoder helps in pretraining the weights of the pixels and the attention block pays more attention to the relevant regions. Our approach achieved 98.27% accuracy when it came to segmenting the cells, it outperformed the state-of-the-art U-N et in terms of traditional parameters.