{"title":"USDL:利用深度学习技术和超声波技术实现低成本医学成像。","authors":"Manish Balamurugan, Kathryn Chung, Venkat Kuppoor, Smruti Mahapatra, Aliaksei Pustavoitau, Amir Manbachi","doi":"10.1115/dmd2020-9109","DOIUrl":null,"url":null,"abstract":"<p><p>In this study, we present USDL, a novel model that employs deep learning algorithms in order to reconstruct and enhance corrupted ultrasound images. We utilize an unsupervised neural network called an autoencoder which works by compressing its input into a latent-space representation and then reconstructing the output from this representation. We trained our model on a dataset that compromises of 15,700 <i>in vivo</i> images of the neck, wrist, elbow, and knee vasculature and compared the quality of the images generated using the structural similarity index (SSIM) and peak to noise ratio (PSNR). In closely simulated conditions, the architecture exhibited an average reconstruction accuracy of 90% as indicated by our SSIM. Our study demonstrates that USDL outperforms state of the art image enhancement and reconstruction techniques in both image quality and computational complexity, while maintaining the architecture efficiency.</p>","PeriodicalId":93509,"journal":{"name":"2020 Design of Medical Devices Conference (DMD 2020). Design of Medical Devices Conferences (2020 : Minneapolis, Minn.)","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2020-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8895229/pdf/","citationCount":"0","resultStr":"{\"title\":\"USDL: Inexpensive Medical Imaging Using Deep Learning Techniques and Ultrasound Technology.\",\"authors\":\"Manish Balamurugan, Kathryn Chung, Venkat Kuppoor, Smruti Mahapatra, Aliaksei Pustavoitau, Amir Manbachi\",\"doi\":\"10.1115/dmd2020-9109\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<p><p>In this study, we present USDL, a novel model that employs deep learning algorithms in order to reconstruct and enhance corrupted ultrasound images. We utilize an unsupervised neural network called an autoencoder which works by compressing its input into a latent-space representation and then reconstructing the output from this representation. We trained our model on a dataset that compromises of 15,700 <i>in vivo</i> images of the neck, wrist, elbow, and knee vasculature and compared the quality of the images generated using the structural similarity index (SSIM) and peak to noise ratio (PSNR). In closely simulated conditions, the architecture exhibited an average reconstruction accuracy of 90% as indicated by our SSIM. Our study demonstrates that USDL outperforms state of the art image enhancement and reconstruction techniques in both image quality and computational complexity, while maintaining the architecture efficiency.</p>\",\"PeriodicalId\":93509,\"journal\":{\"name\":\"2020 Design of Medical Devices Conference (DMD 2020). Design of Medical Devices Conferences (2020 : Minneapolis, Minn.)\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2020-04-01\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8895229/pdf/\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"2020 Design of Medical Devices Conference (DMD 2020). Design of Medical Devices Conferences (2020 : Minneapolis, Minn.)\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1115/dmd2020-9109\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"2020/7/27 0:00:00\",\"PubModel\":\"Epub\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"2020 Design of Medical Devices Conference (DMD 2020). Design of Medical Devices Conferences (2020 : Minneapolis, Minn.)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1115/dmd2020-9109","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"2020/7/27 0:00:00","PubModel":"Epub","JCR":"","JCRName":"","Score":null,"Total":0}
USDL: Inexpensive Medical Imaging Using Deep Learning Techniques and Ultrasound Technology.
In this study, we present USDL, a novel model that employs deep learning algorithms in order to reconstruct and enhance corrupted ultrasound images. We utilize an unsupervised neural network called an autoencoder which works by compressing its input into a latent-space representation and then reconstructing the output from this representation. We trained our model on a dataset that compromises of 15,700 in vivo images of the neck, wrist, elbow, and knee vasculature and compared the quality of the images generated using the structural similarity index (SSIM) and peak to noise ratio (PSNR). In closely simulated conditions, the architecture exhibited an average reconstruction accuracy of 90% as indicated by our SSIM. Our study demonstrates that USDL outperforms state of the art image enhancement and reconstruction techniques in both image quality and computational complexity, while maintaining the architecture efficiency.