AI Ultrasound: A Rising Star in the Field of AI Medical Imaging

In various subfields of artificial intelligence (AI) medical imaging, AI ultrasound has received far less attention than AI CT imaging. In early 2020, the FDA approved the first AI-assisted ultrasound diagnostic software, Caption Guidance, marking a major breakthrough in the field of AI ultrasound and attracting more attention as a result.

Improving Diagnostic Accessibility

Ultrasound examination is a safe and inexpensive medical diagnostic method, but its usage rate in actual clinical practice is not high. Currently, there are approximately 50 million doctors worldwide, but only 2% of them possess ultrasound scanning skills. In addition, according to statistics from the Chinese Medical Equipment Association, China has a considerable number of ultrasound devices, but the distribution among hospitals at different levels is uneven. By the end of April 2018, 2,427 tertiary hospitals in China owned 24,270 color ultrasound devices, averaging 10 devices per hospital; while secondary and primary hospitals averaged 5 and 1 color ultrasound devices respectively, a significant disparity.

Ultrasound examination has the characteristics of being radiation-free and allowing for repeatable diagnosis. With technological advancements, the cost of ultrasound examination is becoming increasingly lower, and the equipment is gradually becoming smaller. Only by lowering the barrier to use can it become a truly accessible and portable diagnostic tool.

Ultrasound diagnosis differs from radiology diagnosis—radiologists can diagnose using static images, while ultrasound doctors need to acquire dynamic images from different sections for real-time diagnosis. The acquisition and diagnosis of ultrasound images are highly dependent on the doctor's experience. AI integrated into ultrasound equipment can help solve two problems: how to better acquire images; and how to better analyze images.

To complete image acquisition and analysis in a short time, AI needs to complete three steps. Taking Caption Guidance as an example, this software first uses AI to guide doctors in image acquisition, allowing non-professional doctors to acquire ultrasound images through AI's real-time guidance; the second step is to use algorithms to find the best images; and the third step is to perform image analysis. Typically, doctors need years of training to interpret and analyze ultrasound images, while Caption Guidance, through deep learning, can automatically measure ejection fraction to assist doctors in assessing patient conditions.

AI makes ultrasound examination simpler and more accessible, while ultrasound is more conducive to leveraging the value of AI. Generally, fields with economies of scale are more likely to create AI value洼地. Compared to CT and MRI, ultrasound has a higher number of applications, therefore, the commercial prospects of AI-assisted ultrasound diagnosis are more attractive.

 "Large and Small" Approaches

The combination of AI and ultrasound is becoming a rising star in the AI imaging sector. In addition to helping with better diagnosis, AI in ultrasound imaging can also perform automated image quality assessment, image standardization processing, image outlining, and automatic measurement. These functions cannot be achieved in ultrasound through a universal solution, therefore, AI ultrasound has taken two completely different routes.

One route is in traditional ultrasound departments, where AI makes large ultrasound equipment more intelligent, transforming ultrasound equipment from merely an imaging product into an intelligent terminal integrating data acquisition, management, and analysis, incorporating deep learning. In 2019, GE Healthcare launched a LOGIQ™ E20 equipped with a cSound+™ image generator in China, which can achieve tissue organ structure identification, intelligent lesion segmentation, and intelligent measurement through image perception, helping doctors to get rid of numerous and cumbersome image optimization and measurement work and focus on clinical diagnosis. This equipment is mainly used in interventional, thyroid, breast, musculoskeletal, pediatric, and cardiac clinical fields to assist clinicians in accurate diagnosis.

Of course, for large-scale equipment, the role of AI is currently just icing on the cake, but it is foreseeable that AI will play an increasingly important role in the future. At the same time, compared to hardware ultrasound equipment, AI software iterates faster, and software and algorithms are expected to become the mainstream research direction in the ultrasound field in the future. Highly digital equipment will generate a large amount of data, and how to interconnect and integrate data is also a key research area.

Another route is to apply it to primary medical care scenarios. There are nearly 900,000 primary medical institutions in China. Among the three aspects of medical care, medicine, and examinations, increasing investment in examinations is essential to solving the structural contradictions of the medical system. Utilizing portable handheld ultrasound devices is a feasible route for ultrasound equipment to empower primary medical institutions.

Currently, the companies mainly focusing on the primary hospital market are startups. These companies mainly apply AI technology to handheld ultrasound devices, and they are more often used by doctors with insufficient ultrasound examination experience. In the past, ultrasound diagnosis relied on professional doctors using their eyes to identify anatomical structures in images, while AI, through intelligent identification, can automatically find the best images and assist in diagnosis, allowing ultrasound examination operators to not be limited to professionally trained doctors, enabling more ordinary doctors in primary medical institutions to perform ultrasound diagnosis.

The "Late Blooming" of the Sector Stems from Technological Barriers

Compared to AI CT imaging, the AI ultrasound sector is not crowded, with only a few startups involved. So, why has AI ultrasound become a late-blooming sector in the field of AI imaging? The main reason is that compared to other imaging fields, AI ultrasound technology is more challenging.

There are three main technical challenges for AI applications in ultrasound:

First, real-time diagnosis must be achieved. Unlike the static images of CT and MRI, ultrasound images are dynamic and real-time. The difficulty of ultrasound examination lies in the simultaneous completion of image acquisition and film reading. The acquisition of CT, MRI, and X-ray images is done by technicians, while film reading is done by radiologists. Ultrasound examination requires simultaneous image acquisition and film reading, which places higher demands on algorithms and computing power for auxiliary diagnostic technologies.

Second, in terms of data, due to the special data browsing, processing, and storage habits of ultrasound images, its image data is more difficult to obtain than CT images, and the size of the database is limited. In addition, the standardization of ultrasound images is low, and the image clarity mainly depends on the ultrasound doctor's operating skills and equipment Model. AI ultrasound requires a strong expert team to clean and analyze this data.

Finally, there are limitations in the algorithm framework. For AI ultrasound companies, whether they can have their own algorithm framework is very important. However, at present, the vast majority of companies use open-source algorithms, and very few companies have their own algorithms. Ultrasound AI is different from radiology AI. To ensure the accuracy and real-time nature of the analysis, it is highly dependent on independently developed algorithm frameworks. If the algorithm is too lengthy, it will lead to slow device processing speed. Ultrasound examination has high real-time requirements, producing dozens or even hundreds of frames of images per second. Without a powerful algorithm, it is impossible to process such a large amount of data.

In addition to the above three points, if AI technology is to be mounted on handheld ultrasound, the problem of computing power limitations also needs to be solved. Because handheld ultrasound devices are much smaller than traditional ultrasound devices, mounting AI software is a great test of AI's computing power. Companies need not only a very accurate Model to analyze ultrasound videos, but also, on this basis, must ensure that the Model can work effectively under the limited resources of a tablet computer or Mobile platform.

In the continuous evolution and advancement of ultrasound technology, both software and hardware capabilities are crucial. Currently, the software system technology barrier of AI ultrasound is higher. In the ultrasound equipment market where hardware is relatively homogeneous, the ability to develop good AI software largely determines the application space of AI ultrasound product hardware. At the same time, for primary hospitals, portable or small-sized devices are more affordable than large-scale equipment. Therefore, AI handheld ultrasound is more suitable for promotion than AI CT and other large-scale equipment. How to achieve intelligent workflow in handheld AI ultrasound to better meet the application needs of primary hospital general practitioners is a key issue that enterprises need to address.

This information originates from the internet. Please contact us for removal if there is any infringement.