Ultrasound (US) can revolutionize and democratize medical imaging if it offers: (1) access for everyone, and (2) excellent image quality, always. MRI offers (2) but is expensive and will thus not likely be able to provide (1). Low-cost US hardware will enable (1) in the future but is not expected to yield the needed breakthrough for (2). Any solution that addresses both will have a huge impact.
The biggest adversaries for US image quality stem from patient- and user-specific factors. I therefore strongly believe that achieving (2) requires closed-loop, goal-directed imaging through systems that actively pursue information gain, in-situ. The ERC stg project “US-ACT” develops this new closed-loop US paradigm, with deep generative models governing beliefs about anatomy and function. This, however, gives rise to a new dilemma: How to embrace high-performant, but complex, closed-loop algorithms to achieve (2), while at the same time lowering cost to achieve (1)? Our recent result shows that US-ACT enables breaking the cost-complexity trade-off, through a new approach that is rooted in the closed-loop proposition itself.
We propose CloudSound, a fully cloud-native closed-loop US imaging paradigm, unlocking access to virtually unlimited computational power, memory, and storage for raw US sensor data. We will move all the computational complexity away from the local device to a remote server. To achieve this with minimal latency and high frame rates, CloudSound introduces a new holistic rate control strategy for the entire transmit-receive scheme, based on algorithms developed in US-ACT. The intrinsic incentive to maximize image quality as well as an “age of information” functional will drive rate control of the full solution under a time-varying information bottleneck: the network channel.
