Machine Learning & Deep Learning: Clinical Applications
Shigeru Kiryu1

1Radiology, International University of Health and Welfare, School of Medicine, Chiba, Japan

Synopsis

This talk will introduce the clinical applications of deep learning becoming a reality. It will also discuss the unique aspects of deep learning in clinical applications along with its limitations.

Overview

Machine learning, particularly the deep learning approach, has attracted a great deal of attention in the field of radiology. Radiology has progressed with the development of techniques unique to the various processes involved, such as image acquisition, reconstruction, evaluation, diagnosis, prediction. Now, deep learning is being applied in each of these processes, and its high performance has a significant impact. A number of recent reports have explored the utility of deep learning in various processes, and it is becoming apparent that deep learning will have specific roles in the practice of radiology. This presentation introduces the clinical applications, including denoising, segmentation, detection, classification, and unique aspects of deep learning in clinical applications along with its limitations.

MR images are degraded by noise generated during image acquisition, and image denoising is a major research topic of MRI. Deep learning has been applied to denoise images in dynamic contrast-enhanced MRI, 3D MRI, and MR spectroscopy. Deep learning denoising is also expected to be applicable to images obtained with high-speed MRI method. Segmentation of organs is a basic image-processing technique for medical images and is one of the common topics for application of deep learning. Deep learning has been applied to images of various organs already. Detection of lesions is a common task for radiologists. The usefulness of machine learning in lesion detection is recognized as a major benefit of artificial intelligence in diagnostic imaging. Compared to conventional machine learning, deep learning accelerates the performance of lesion detection in a shorter time. Unlike conventional machine learning, deep learning does not require human-defined classification rules. Classification using the deep learning approach has been applied in MRI for many diseases. Classification using deep learning, along with transfer learning, is expected to improve diagnostic performance.

Deep learning has also been applied in unique ways. Using a deep learning approach, pseudo-CT images can be generated from MR images for PET/MR attenuation correction. Synthetic images created using generative adversarial networks serve as data augmentation to improve diagnostic performance.

It is important to be aware of the limitations of deep learning for its clinical application. The features and calculations used by deep learning to perform its processing are challenging for humans to interpret. This is known as the ‘black box’ problem, and ‘Explainable AI’ is being developed to deal with this problem. Also, similar to conventional machine learning, overfitting is a common problem of deep learning. Large amounts of data with valid reference labels are necessary to solve this problem, and this makes research using deep learning laborious.

Acknowledgements

None.

References

  1. Soffer S, Ben-Cohen A, Shimon O, Amitai MM, Greenspan H, Klang E. Convolutional Neural Networks for Radiologic Images: A Radiologist's Guide. Radiology. 2019;290:590-606.
  2. Selvikvåg Lundervold A, Lundervold A. An overview of deep learning in medical imaging focusing on MRI. Z Med Phys. 2018: S0939-3889(18)30118-1.
  3. Benou A, Veksler R, Friedman A, Riklin Raviv T. Ensemble of expert deep neural networks for spatiotemporal denoising of contrast-enhanced MRI sequences. Med Image Anal 2017;42:145–159.
  4. Jiang D, Dou W, Vosters L, Xu X, Sun Y, Tan T. Denoising of 3D magnetic resonance images with multi‑channel residual learning of convolutional neural network. Jpn J Radiol. 2018;36:566-574.
  5. Bien N, Rajpurkar P, Ball RL, Irvin J, Park AK, Jones E, et al. AI-assisted diagnosis for knee MR: Development and retrospective validation. PLoS Med. 2018;15:e1002699.
  6. Yasaka K, Akai H, Kunimatsu A, Abe O, Kiryu S. Liver Fibrosis: Deep Convolutional Neural Network for Staging by Using Gadoxetic Acid-enhanced Hepatobiliary Phase MR Images. Radiology. 2018;287:146-155.
  7. Liu F, Jang H, Kijowski R, Bradshaw T, McMillan AB. Deep Learning MR Imaging-based Attenuation Correction for PET/MR Imaging. Radiology. 2018;286:676-684.
  8. Frid-Adar, M, Klang, E, Amitai, M, Goldberger, J, Greenspan, H. Synthetic data augmentation using GAN for improved liver lesion classification. arXiv:1801.02385.
Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)