SHF:Medium:Collaborative Research: Advanced Architectures for Hand-held 3D Ultrasound Thomas Wenisch, EECS, Univ. of Michigan; Chaitali Chakrabarti, ECEE, Arizona State Univ. J. Brian Fowlkes, Radiology, Univ. of Michigan; Jonathan Rubin, Radiology, Univ. of Michigan Much as every medical professional listens beneath the skin with a stethoscope today, we foresee a time when hand-held medical imaging will become as ubiquitous; peering under the skin using a hand-held imaging device. Ultrasound is an ideal platform: (1) a hand-held probe imaging deep inside the patient; and (2) the energy required in each ultrasound pulse is miniscule compared to magnetic resonance or Xray imaging, making battery operation feasible. Industry has already recognized these advantages with several hand-held ultrasound devices marketed today. However, current devices produce a low-resolution two-dimensional view of a slice of the patients organs and fail to produce the high imaging quality of non-portable systems. Real-time 3D ultrasound drastically improves system ease-of-use, and has already demonstrated improvements in diagnostic efficiency. Moreover, direct 3D image acquisition enables new diagnostic capabilities that are difficult or impossible to accomplish with 2D or 2.5D (3D images reconstructed from 2D slices), for example, accurate volumetric blood flow measurement or 3D sheer wave tracking for tissue elastography. However, the extreme computational requirements (and associated power requirements) of image formation for a large 3D system have, to date, precluded hand-held 3D-capable devices. The multiplicative increases of both the transducer array dimensions and imaging volume combine to increase raw computational requirements by nearly 5000 over comparable 2D imaging. Because it is in close contact with human skin, an ultrasound scan head must operate within a tight power budget (about 5W) to maintain safe temperatures. The demanding power budget mandates innovation in both signal processing algorithms and hardware architecture, motivating a hardware/software co-designed approach. In preliminary work, the PI team has proposed the Sonic Millip3De, a new hardware architecture for 3D hand-held ultrasound that leverages co-design of hardware and beamforming algorithms, 3D die stacking, massive parallelism, and streaming data flow, to enable high-resolution synthetic aperture 3D ultrasound imaging in a single, low-power chip. Based on synthesis of RTL-level hardware designs, Sonic Millipe3De is projected to achieve a sub-5W power target by the 16nm technology node. However, the preliminary design is tuned only for abdominal imaging at low frame rates (~1 Hz) and lacks the computational capability and programmability/flexibility for advanced ultrasound applications. Intellectual Merit. This project comprises two key thrusts: (1) Algorithmic Innovationalgorithm and hardware extensions to Sonic Millip3De motivated by specific clinical applications, and (2) Closing the Evaluation Loopdemonstration of an FPGA-based prototype reconstructing images of physical phantoms acquired with commercial scan-heads. The effort focuses on three application areas: (i) aberration correction, motion composition, and separable beamforming to enhance image quality and manage computational requirements of general imaging applications, such as abdominal imaging; (ii) speckle tracking of 3D sheer waves to enable tissue elastography, including novel applications to the diagnosis of chronic obstructive pulmonary disorder; and (iii) high-frame-rate 3D Doppler flow tracking for cardiac applications, with the ambitious objective of capturing complete flow fields to enable direct computation of cardiovascular pressure drops. Broader Impacts. This effort seeks to enable ubiquitous availability of advanced 3D ultrasound imaging. Hand-held imaging is not only a matter of convenience; moving the imaging device to the patient (rather than a critical patient to the radiology lab) has been shown to improve clinical outcomes. Moreover, the portability of hand-held systems can make advanced imaging available to traditionally underserved populations in the rural and developing world. The PI Team brings together experts in signal processing, computer architecture, medical ultrasound, and clinical radiology. Through hardware and algorithm codesign, the team can leapfrog industrial efforts to implement ultrasound systems with commodity processors. The targeted medical applications span clinically proven diagnostic methods that can be Project Summary Pag e 2 shifted from the radiology lab to the field and new applications that rely on the algorithmic innovations we propose. Hence, the impacts of this project go beyond hardware and computing systems, facilitating future medical research. The PIs have a track-record of engaging women in computing, and the immediate societal relevance of the proposed work is seen as a key asset to enhancing those recruiting and retention efforts.
|Effective start/end date||6/1/14 → 5/31/19|
- National Science Foundation (NSF): $400,000.00