The future of medical imaging, AI empowers faster and more accurate judgments

Figure 1: Infervision artificial intelligence-assisted diagnosis improves the overall efficiency of the radiology department

(Image source: Infervision)

Fortunately, AI can play a huge role in image recognition and image processing. According to newly published research results in The Lancet Gastroenterology and Hepatology, artificial intelligence-assisted endoscopy can reduce the risk of missing gastric tumors by nearly 80%. Through efficient AI algorithms combined with high-performance terminal processing chips, doctors can help doctors achieve faster, more accurate and high-volume medical imaging diagnosis.

AI medical imaging can be divided into X-ray, ultrasound, CT, MRI and clinical diagnostic endoscopy. Different diagnostic methods correspond to different disease characteristics and onset locations, so the processing requirements for AI are also different.

Figure 2: AI medical image classification

(Source: Leopard)

Figure 3: Market expectations for AI medical imaging-assisted diagnosis

(Source: IDTechEx Research)

The application of AI medical imaging has already achieved a closed loop of the value chain. Next, it will move towards large-scale popularization, which requires the realization of two-wheel drive in technology and clinical aspects. Qian Dahong, a professor at the School of Biomedical Engineering at Shanghai Jiao Tong University, believes that the development of AI-assisted diagnosis has two important elements, one is clinical drive, and the other is core technology drive. To integrate AI-assisted diagnosis into doctors' workflow and become a strong clinical need, this is the key to implementation.

At the algorithm level of core technology, multi-modal input is the focus of current exploration in the industry, and it is equally important to achieve unsupervised artificial intelligence learning. But the so-called key to technology implementation not only refers to AI algorithm technology, but more importantly, the combination of software and hardware. For different AI-assisted diagnostic medical devices, it is necessary to have a chip platform that matches the scenario and algorithm to truly amplify the capabilities of AI and enable AI-assisted diagnosis on the device side and in the clinic.

So what are the different requirements for chip platforms for different medical imaging diagnoses? This should be viewed in conjunction with the technological development of respective diagnostic equipment.

The new generation of ultrasonic imaging is a change in thinking mode. As a well-known manufacturer in the field of FPGA, Xilinx's products can upgrade ultrasonic detection from normal sequential acquisition to parallel acquisition of the entire plane. Xilinx's Versal AI series not only provides a beamformer running synthetic aperture or plane wave algorithms, but also provides 128 operating elements and 200 lines of resolution on a single chip. Through the Xilinx ACAP platform, ultrasonic testing equipment can achieve scan rates of hundreds to thousands of frames per second from cardiac imaging to abdominal imaging to the detection of any small components. In addition, Xilinx ZYNQ-Ultrascale+SoC technology can also implement multi-beam forming in miniaturized ultrasound equipment, integrating up to 8 parallel beamformers in handheld devices.

Figure 4: Ultrasonic detection solution of Xilinx Versal AI series

(Image source: Xilinx)

Figure 5: Xilinx’s dual 4K endoscope design example

(Image source: Xilinx)

Regardless of the above-mentioned medical imaging equipment, it contains multiple subsystems such as end-side image acquisition, pre-processing calculation, cloud data processing and recording. The overall complexity of the AI ​​medical imaging system is high, the hardware solutions are difficult to unify, and the coupling between top-level algorithm development and underlying hardware implementation is difficult. In order to solve this problem, Xilinx not only provides many AI acceleration chip platforms, but also cooperates with algorithm API vendors and cloud service vendors to provide a complete AI medical imaging solution integrating edge to cloud, thereby assisting the development of different types of AI imaging systems.

Figure 6: Clarius’ ultra-portable ultrasound device designed with Xilinx Zynq

(Image source: Xilinx)

How to cater to the development needs of AI medical diagnosis and quickly develop AI medical imaging solutions that can truly empower clinical applications? The next two evaluation kits may help you get started.

The first is the Xilinx Zynq UltraScale+ MPSoC ZCU102 evaluation kit , the part number on Mouser Electronics is EK-U1-ZCU102-G-ED . The kit uses a Zynq UltraScale+ MPSoC device with quad-core ARM Cortex-A53, dual-core Cortex-R5 and Mali-400 MP2 graphics processing unit based on Xilinx's 16nm FinFET programmable logic structure.

Figure 7: Xilinx Zynq UltraScale+ MPSoC ZCU102 Evaluation Board

(Source: Mouser Electronics)

The evaluation kit integrates 4GB 64-bit DDR4 SODIMM and 512MB DDR4 component memory, and provides PCIe, USB3, display port and SATA interfaces. The ZCU102 evaluation kit has a wealth of onboard resources and interfaces. Together with Xilnx's Vivado software development platform, it can quickly help developers complete basic peripheral control, create logic designs, and package them into IP. IP simulation verification can be performed directly on the board to accelerate Prototype development for customers.

The second development board we recommend is the Xilinx Versal AI Core series VCK190 evaluation kit . The product number on Mouser Electronics is EK-VCK190-G-ED . The VCK190 evaluation kit is designed for designs requiring high-throughput AI inference and signal processing. The kit has 100 times the computing power of server-class CPUs, offers a variety of connectivity options and standardized development processes. For AI image computing, the kit has a PCIe Gen4 interface to support high computing performance and data handling, and is equipped with an HDMI interface for video processing applications to facilitate direct connection of screens and sensors.

Figure 8: Xilinx Versal AI Core Series VCK190 Evaluation Kit

(Source: Mouser Electronics)

Driven by AI, the medical imaging industry will undergo tremendous changes. The results of a questionnaire survey conducted among 332 imaging professionals in 24 provinces and cities in China in 2022 showed that after using AI products, the proportion of departments that can complete more than 50 cases per day on average increased from 14.76% to 34.04%; 70.48% of departments There are procurement needs for AI disease solutions. 2023 may become the first year of the explosion of smart medical care.

We will see that as the medical insurance system improves and the clinical application rate of AI imaging equipment increases, the AI ​​medical imaging track will accelerate into a cycle of positive value feedback. The black and white combination of black technology and white coats will provide us with Bringing a more intelligent and efficient medical treatment experience.