Towards Clinical Application of Artificial Intelligence in Ultrasound Imaging

Abstract

Artificial intelligence (AI) is being increasingly adopted in medical research and applications. Medical AI devices have continuously been approved by the Food and Drug Administration in the United States and the responsible institutions of other countries. Ultrasound (US) imaging is commonly used in an extensive range of medical fields. However, AI-based US imaging analysis and its clinical implementation have not progressed steadily compared to other medical imaging modalities. The characteristic issues of US imaging owing to its manual operation and acoustic shadows cause difficulties in image quality control. In this review, we would like to introduce the global trends of medical AI research in US imaging from both clinical and basic perspectives. We also discuss US image preprocessing, ingenious algorithms that are suitable for US imaging analysis, AI explainability for obtaining informed consent, the approval process of medical AI devices, and future perspectives towards the clinical application of AI-based US diagnostic support technologies.

Keywords: artificial intelligence; classification; deep learning; detection; explainability; machine learning; preprocessing; segmentation; ultrasound imaging.

Figure 1

Acoustic shadow detection: (a) The red areas represent the segmented acoustic shadows using the semi-supervised approach [43]. (b) As a candidate for clinical application, examiners can evaluate whether the current acquired US imaging is suitable for diagnosis in real time. In the case of low image quality, rescanning can be performed in the same examination time. This application may improve the workflow of examiners and reduce the patient burden.

Figure 2

Fundamental algorithms generally used in US imaging analysis. (a) Image classification of whether the fetal US image contains a diagnostically useful cross-section such as a four-chamber view (4CV). (b) Detection of the fetal heart for evaluation of fetal heart structure. (c) Segmentation of the boundaries or regions of the fetal heart to measure the fetal cardiac index such as cardiothoracic area ratio (CTAR).

Figure 3

Use of time-series information to reduce noisy artifacts and to perform accurate segmentation in US videos. CSC employs the time-series information of US videos and specific section information to calibrate the output of U-Net [51].

Figure 4

Possible techniques for AI explainability. The cardiac substructures were detected with colored bounding boxes in a three-vessel trachea view in (a) a normal case, and (b) a tetralogy of Fallot (TOF) case. (c) An image of the class-specific heatmap indicates the discriminative regions of the image that caused the particular class activity of interest. (d) Barcode-like timeline in a TOF case. The vertical axis represents the 18 selected substructures and the horizontal axis represents the examination timeline in the rightward direction. A probability of ≥0.01 was set as well-detected and is indicated as the blue bar, whereas <0.01 was set as non-detected and is indicated by the gray bar in each frame. The pulmonary artery was not detected (red dotted box).