Live countdown: AI acceleration helps medical imaging technology enter the market faster
Vitis library accelerates advanced medical imaging on Versal devices with AI engine support, delivering frame rates up to 1000 FPS with low latency
AMD launches the first medical imaging libraries for the Vitis™ 2022.2 unified software environment. Customers can accelerate their time to market for medical imaging products using a range of available options from L1 building blocks to L3 complete ultrasound beamformer designs. Our software libraries, documentation, and example designs demonstrate how to build this design using Versal™ devices with AI engines. Any medical imaging device manufacturer can choose to quickly deploy its own algorithms using L1 basic building blocks and high-level programming languages.
For ultrasound equipment manufacturers, the L2 library can be used to deploy their own algorithms using high-level programming languages for rapid time-to-market. The tool suite in our L2 library contains all the functions and examples on how to build a custom high-performance beamformer, while the L3 library provides a ready-made application that supports a fully working beamformer.
We will introduce you to this newly released medical imaging library in detail during the live broadcast on March 23, and show you how to use the Versal adaptive computing acceleration platform with AI computing engine to build this through some reference designs. Class design plan.
Vitis™ Medical Imaging Library—Ultrasound Reference Design
The Vitis Accelerated Medical Imaging Library is primarily used for next-generation UltraFast™ imaging reference design examples in the L3 library. This reference design represents a paradigm shift in medical imaging, where we can go from normal sequential acquisition to parallel acquisition across an entire plane, using spherical or plane waves. This helps our customers create optimal focus wherever an image is captured, delivering ultra-high quality 3D/4D images and enabling thousands of images per second.
AMD ACAP platforms, such as the AI Core and AI Edge series devices based on the AI engine, not only provide a beamformer running synthetic aperture or plane wave algorithms , but also provide 128 working elements and 200 lines of resolution on a single chip , and can achieve scan rates of hundreds to thousands of frames per second on anything from cardiac imaging to abdominal imaging to small-site imaging.
Vitis™ AI Engine Medical Image Library—Function and Orchestration
The Vitis AI Engine Medical Imaging Library is a series of configurable components that can be used to develop various medical applications through the Versal AI Engine. These open source libraries are mainly targeted at target DSP applications, including ultrasound beam forming, CT image reconstruction, MRI image reconstruction using 2D-FFT, gradient processor control of MRI magnets, image processing of sampled/digitized data of X-ray input, And ECG etc.
The current version provides basic level 1 building block support for a variety of medical imaging applications for a complete reference design for building advanced ultrasound beamformers. It contains:
Supports BLAS (similar to Numpy)
Support basic DSP (such as the windowing function of diffraction image control method)
Interpolation (spline interpolation)
A complete medical ultrasound beamformer that uses all of the above features
There are three different levels of these libraries:
L1 —The lowest level abstraction (consisting of simple BLAS operations). These operations are C++ implementations of the open source NumPy library. The L1 library supports both arithmetic operations and some vector processing operations.
L2 — Functional unit of the beamformer (available by combining the L1 library). The Level 2 ultrasound library consists of the mathematical components required for beamforming of RF data. Compared to the L1 API, these components are AIE graphics and no longer a single kernel. Therefore, the level of abstraction is increased compared to the L1 level.
L3 —Full beamforming design. The last level of abstraction is a complete beamformer, consisting of units provided at L2 level. Three types of beamformers are available: Synthetic Aperture (SA), Plane Wave (PW) and ScanLine.
Speakers
Trevor Weng
AMD Industrial and Vision Systems Architect
Trevor Weng is currently a system architect at AMD, responsible for industrial and visual market strategy, business development and system solutions in Greater China. He has over 16 years of experience in the medical, communications, video, embedded and semiconductor industries.
Live broadcast reservation
* By submitting the corresponding personal information, you agree to disclose your data to AMD. Due to the organization of the event and AMD's corresponding internal management and system operation needs, your information may be transferred to AMD Inc. in the United States, and according to the published Privacy Policy for processing.
* Unless otherwise stated, the event organizer is not responsible for the cancellation of the event due to special reasons or the inability to enter the live broadcast room to participate in the event due to failure to meet the requirements after successful registration.
京公网安备 11010802033920号