Sec. Artificial Intelligence in Radiology
Volume 2 - 2022 | https://doi.org/10.3389/fradi.2022.1097533
Editorial: Advances in deep learning methods for medical image analysis
- 1Department of Artificial Intelligence, Korea University, Seoul, South Korea
- 2Department of Brain and Cognitive Engineering, Korea University, Seoul, South Korea
- 3Department of Radiology and BRIC, University of North Carolina at Chapel Hill, Chapel Hill, NC, United States
- 4Shanghai United Imaging Intelligence Co., Ltd., Shanghai, China
- 5School of Computer Science & Engineering, Kyungpook National University, Daegu, South Korea
Editorial on the Research Topic
Advances in deep learning methods for medical image analysis
The rapid development of artificial intelligence (AI) technology is leading many innovations in the medical field and is playing a major role in establishing objective, consistent, and efficient medical environments with large-scale data. Deep learning represented by convolutional neural networks has achieved remarkable performance improvement in medical image processing fields such as image segmentation, registration, and enhancement. Furthermore, AI technology with deep learning is pioneering medical applications, such as lesion detection, differential diagnosis, disease prognosis, and surgical planning. More advanced AI technologies, such as transformers with self-attention mechanisms, allowing for learning global dependencies, have been widely applied, which further enhanced the capability of deep learning to analyze medical images. However, despite the remarkable advances in deep learning, many challenges remain. For example, when training data are biased or incomplete, deep learning models may fail to achieve the good generalization capability required to solve real-world problems. In addition, the limitations of deep learning models in interpreting results, and misunderstandings of their intended uses and hypotheses make it difficult for AI to gain trust in healthcare settings. In this regard, disease-specific neural networks, generalized learning methods, high-quality training data, and external evaluation based on testable hypotheses can ensure the reliability of medical AI technologies for humans (1, 2).
In medical image analysis, image segmentation has a role to accurately delineate organs or lesions from medical images, and to provide quantitative information of target objects. It is important for image segmentation to understand the various patterns of target objects in medical images, and to show robust performance against large variations in image quality. In this special issue on “Advances in Deep Learning Methods for Medical Image Analysis”, Zhang et al. proposed a brain tumor segmentation approach that ensembles three segmentation networks based on U-Net that receive different MR modalities (such as T1-weighted, T2-weighted and brain parcellation) as input, and also used a post-processing strategy to distinguish small enhanced tumors to reduce false-positives Zhang et al. Kuijawa et al. proposed ensemble-based and multi-task neural networks to perform Koos grading for Vestibular Schwannoma (VS) using brain tumor segmentation Kuijawa et al. In particular, the performance of Koos grading was improved by majority voting between an end-to-end neural network with multi-modal MR images and random forest models using radiomic features extracted from the segmentation of the brain structures related to VS Kuijawa et al. Bouget et al. showed the ability of U-Net-based architecture with an attention mechanism in segmenting Meningiomas (a type of brain tumor) with 98% accuracy for brain lesions larger than 3 ml Bouget et al. Aboutalebi et al. introduced a multi-scale encoder-decoder self-attention mechanism tailored for medical image analysis. They achieved higher performance in identifying SARS-CoV-2 in chest x-rays than conventional attention-based U-Net architectures Aboutalebi et al. These studies show that neural network architectures can improve generalization by learning and discovering varied disease-related pattern representations in medical images. At the same time, the fusion of multiple models helps provide diverse information required in the diagnostic process.
Although there is no golden method that can solve all the challenges faced in natural environments, it seems evident that deep learning technology is augmenting human ability in medical image analysis and raising new approaches to diagnosis and treatment in clinical settings. In the next few years, we may observe even more rapid changes in the medical environment led by AI with deep learning, further requiring more reliable AI technologies. Researchers will continue to investigate advanced AI technologies, just as transformers are now widely applied to the medical imaging field, and more advanced algorithms are expected.
The authors contributed equally for this editorial. All authors contributed to the article and approved the submitted version.
Conflict of interest
XC is employed by Shanghai United Imaging Intelligence Co., Ltd. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
1. Rotner J, Hodge R, Danley L. Five AI Fails and How We Can Learn from Them, Jonathan Rotner, Ron Hodge, Lura Danley (Ph.D.), MITRE (2021). Available from: https://www.mitre.org/news-insights/publication/five-ai-fails-and-how-we-can-learn-from-them
2. Quinn TP, Senadeera M, Jacobs S, Coghlan S, Le V. Trust and medical AI: the challenges we face and the expertise needed to overcome them. J Am Med Inform Assoc. (2021) 28(4):890–4. doi: 10.1093/jamia/ocaa268
Keywords: editorial, deep learning, artificial intelligence, medical image analysis, neural network
Citation: Suk H, Liu M, Cao X and Kim J (2023) Editorial: Advances in deep learning methods for medical image analysis. Front. Radio 2:1097533. doi: 10.3389/fradi.2022.1097533
Received: 14 November 2022; Accepted: 6 December 2022;
Published: 6 January 2023.
Edited and Reviewed by: Syed Anwar, University of Engineering and Technology Taxila, Pakistan
© 2023 Suk, Liu, Cao and Kim. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Heung-Il Suk firstname.lastname@example.org Mingxia Liu email@example.com Xiaohuan Cao firstname.lastname@example.org Jaeil Kim email@example.com
Specialty Section: This article was submitted to Artificial Intelligence in Radiology, a section of the journal Frontiers in Radiology