The collection and analysis of large volumes of information, big data, has reached the ultrasound imaging market and its application will fundamentally change the industry. Rising healthcare costs and consumerism have driven the demand for more efficient and effective medical imaging. Meanwhile, digitalization of medical imaging equipment has enabled the storage and analysis of an abundance of digital medical data. Artificial intelligence (AI) has been trained on this myriad of data and can be useful for ultrasound to improve both scan efficiency and diagnosis. Despite the large potential upside, the application of big data to ultrasound may be occurring too quickly, resulting in unintended consequences, primarily surrounding patient data security.
AI will play a pivotal role in making ultrasound more efficient and more accurate. Currently, AI is primarily applied to make scans quicker and easier for the practitioner, by automating some elements of the scan. For example:
Philip Healthcare’s “Anatomical Intelligence” tool automatically measures and quantifies anatomical features in an ultrasound scan.
GE Healthcare’s “Scan Coach” application guides users through a scan.
Konica Minolta’s “Adaptive Imaging” software anticipates the user’s needs and adjusts settings accordingly.
These algorithms save healthcare providers time by eliminating repetitive tasks and assist in diagnosis, improving cost efficiency and patient outcomes. This is most notable in understaffed facilities and with undertrained practitioners.
While in most cases big data is used for automation, the long-term and greater impact would be its application for diagnostics. AI companies, like DiA and Koios, have impressed the medical imaging community with machine learning software that use big data to aid in cardiac and breast diagnostics, respectively.
DiA Imaging Analysis has made a significant impact on cardiovascular ultrasound with its LVivo Toolbox, a set of AI-powered applications that aid in ultrasound interpretation. DiA’s algorithms can be used for ejection fraction (EF) measurements, heart disease detection, and identifying abnormalities during coronary events. In recent years, DiA has rapidly expanded its presence in the ultrasound market through partnerships with several medical imaging manufacturers as well as technology giants, Google Cloud and IBM Watson Health.
Koios Medical recently received FDA approval for its DS Breast 2.0, a machine learning algorithm trained to detect breast cancer from breast ultrasounds. This technology is available on GE’s Logiq E10. Koios is currently working to develop similar AI algorithms to help diagnose thyroid, skin, and prostate cancers.
AI is especially important in the quickly growing segment of point-of-care (POC) ultrasound, where scans are performed outside of the traditional hospital setting. Efficiency and ease-of-use are crucial factors in fast-paced POC environments, such as emergency rooms and ambulances, and when end-users lack advanced ultrasound training. Many ultrasound systems designed for POC are equipped with AI applications, including:
GE Healthcare’s Venue Go and Vscan
Butterfly Network’s Butterfly iQ
Several of Clarius’ handheld scanners including the C3 and PA
Additionally, Fujifilm Sonosite, Fujifilm Medical System’s ultrasound business unit that specializes in POC, has signed collaborative agreements with Partners Healthcare and AI2 Incubator to further develop AI for POC ultrasound.
Despite remarkable innovations, AI for ultrasound faces several challenges. These challenges include the lack of image quality and consistency in acquisition for ultrasound compared to other medical imaging modalities and a disconnect between AI research and healthcare application. The extent that AI will alter ultrasound is unknown, but collaborations and product launches (illustrated in the timeline below) are an indicator of the direction that AI for ultrasound is heading.
While many stakeholders in the medical imaging community embrace big data for its long-term benefits, others stress its concerning implications, principally surrounding patient data security. Healthcare providers store more data than ever before, and a breach of a healthcare network would expose the medical records of countless patients.
Connected devices such as ultrasound systems, are often one of the most vulnerable points of an information technology (IT) network to hack. Ultrasound technology has made enormous advancements, however developments in cybersecurity has lagged. While POC ultrasound expands the presence of medical imaging, it also increases a network infrastructures exposure to infiltration by removing devices from the secure IT environment of most hospitals. The ECRI Institute identified POC ultrasound as one of the “Top 10 Health Technology Hazards for 2020” because of concerns including data security.
Much of the healthcare community believes that cybersecurity oversight does not adequately ensure patient data security, and thus is anticipating the FDA to crackdown on ultrasound use and data archiving, especially in POC settings. Such a crackdown could drastically hinder the growth of the POC ultrasound market and AI, as developers will need to focus their attention on cybersecurity instead on automation and diagnostics. Device manufacturers have developed and tout their cybersecurity capabilities, such as GE Healthcare’s ultrasound-specific data security feature SonoDefense, to ease some of the hesitation among healthcare providers to adopt innovative ultrasound technology.
With all the excitement surrounding big data, it is easy to get carried away focusing on its major upside. Big data can be harnessed through ultrasound to drastically increase access to fast and accurate medical imaging. Nonetheless, it is important that the medical imaging community takes its time to not only ask how efficient or how effective big data can make ultrasound, but also at what expense?