Compact modular open platform for low-cost ultrasound imaging

Abstract

Ultrasound open platforms (OPs) offer the flexibility and reconfigurability required for developing novel ultrasound techniques. However, they are generally large and expensive. This limits their use in primary care and low-resource settings. In this thesis, we propose a modular OP to address these challenges.

The modular architecture of the proposed OP enables trading off size and cost to meet performance targets. The OP is composed of five core modules and is suitable for a wide range of ultrasound applications. The intrinsic modularity of the OP enables the extension of its performance by adding extension modules. Furthermore, the OP features a digital back-end controller that is designed specifically for ultrasound imaging and implemented on an Artix-7 FPGA.

In particular, the OP is capable of generating arbitrary waveforms with a maximum frequency of 20 MHz and a maximum amplitude of $\pm100$ V. It can also generate precise delays with a maximum error of 200 ps. Additionally, the OP enables varying the gain from 12 dB to 51 dB. It also has a maximum ADC resolution of 12 bits, and a maximum conversion rate of 100 MSPS.

Results from various phantoms and in vivo experiments demonstrate that the OP yields high-quality images and enables the detection of anatomical structures in small animals. Additionally, the OP has been deployed in two research projects, where it was shown to fulfil the requirements of both projects at a fraction of the cost and size of commercial OPs.

Finally, a novel analogue accelerator for the training of AI models using continuous-time stochastic gradient descent is proposed. The accelerator is intended for applications that demand privacy and on-device training, such as ultrasound imaging. The accelerator was simulated in AMS 0.35 um technology. Results show that the accelerator offers excellent accuracy, with at least 8 bits of resolution.