Deep Learning in the Echocardiography Workflow: Challenges and Opportunities
Doctoral thesis
Permanent lenke
https://hdl.handle.net/11250/3108433Utgivelsesdato
2023Metadata
Vis full innførselSamlinger
Sammendrag
Echocardiography is the most widely used imaging technique for cardiac imaging due to its availability, low-cost and real-time capabilities. With numerous indicators of heart function, it now enables precise diagnosis of heart diseases. However, echocardiography is primarily practiced within hospitals by expert operators and has some limitations, such as laborintensive interpretations and high operator dependence. On the other hand, portable echocardiography devices have entered the market, offering new possibilities for point-of-care applications and increasing access to echocardiographic assessment in developing countries. Nonetheless, new operators using portable devices may have limited expertise compared to hospital residents, which could potentially hinder the full utilization of portable devices in the field.
The overall goal of this thesis is to investigate the use of deep learning (DL) techniques to address the limitations of echocardiography, from both the hospital and point-of-care perspectives. We first investigated the challenges when using DL to automate the interpretation of echocardiographic images. Secondly, the potential of using DL to assist operators by guiding them during scanning is explored. Finally, a clinical study of the developed guiding application is proposed.
Results indicate that using DL to automate echocardiographic image interpretation is challenging, and that the development of clinically valuable DL models requires expertise in ultrasound, clinical, and statistical domains.
These findings and the proposed mitigation solutions may have implications for the development and evaluation of future DL tools for interpretation of echocardiographic images. Further, the proposed real-time application for guiding operators demonstrates that DL offers an excellent opportunity to facilitate and improve image acquisition. The clinical study show that the developed method is beneficial in the hospital setup, with the possibility to reduce the operator dependence of echocardiography. Further studies should investigate the value of real-time guiding for non-expert users in point-of-care settings.
Består av
Pasdeloup, David; Østvik,Andreas; Olaisen, Sindre H.; Skogvoll, Eirik; Dalen,Håvard; Løvstakken,Lasse. Challenges and Strategies for Automatic Measurements with Deep Learning in Cardiovascular ImagingPasdeloup, David Francis Pierre; Olaisen, Sindre Hellum; Østvik, Andreas; Sæbø, Sigbjørn; Pettersen, Håkon Neergaard; Holte, Espen; Grenne, Bjørnar; Stølen, Stian Bergseng; Smistad, Erik; Aase, Svein Arne; Dalen, Håvard; Løvstakken, Lasse. Real-Time Echocardiography Guidance for Optimized Apical Standard Views. Ultrasound in Medicine and Biology 2023 ;Volum 49.(1) s. 333-346 https://doi.org/10.1016/j.ultrasmedbio.2022.09.006 This is an Open Access article distributed under the terms of the Creative Commons Attribution License CC BY
Sæbø, Sigbjørn; Pasdeloup, David; Pettersen, Håkon N.; Smistad,Erik; Østvik, Andreas; Olaisen, Sindre H.; Stølen, Stian B.; Grenne, Bjørnar L: Holte,Espen; Løvstakken,Lasse Dalen,Håvard. Real-time Guiding by Deep Learning of Experienced Operators to Improve Standardization of Echocardiographic Acquisitions European Heart Journal - Imaging Methods and Practice, Volume 1, Issue 2, September 2023, qyad040, https://doi.org/10.1093/ehjimp/qyad040 This is an Open Access article distributed under the terms of the Creative Commons Attribution License CC BY