Elliptic Labs (Norway)

Bayesian approaches to Neural Architecture Search (NAS) offer several benefits when compared to typical grid-search methods. Marginalization of a joint posterior distribution of network parameters as well as network architecture specification allows for a principled analysis of predictive and model uncertainty. In this paper, we propose a method of applying Reversible Jump Markov Chain Monte Carlo (RJMCMC) as inference over a distribution of neural network depths. We refer to these models as Reversible Jump Bayesian Neural Networks (RJBNN), and show that a tenable acceptance probability for across-dimension proposals can be achieved through the use of a composite sampler featuring the Hamiltonian Monte Carlo (HMC) based No U-Turn Sampler (NUTS) algorithm. We compare the results of an RJBNN model with standard Bayesian Neural Networks (BNN) of comparable architecture sizes on a regression task for the Boston Housing dataset.

External beam radiation therapy (EBRT) is included in the treatment regimen of the majority of cancer patients. With the proliferation of hypofractionated radiotherapy treatment regimens, such as stereotactic body radiation therapy (SBRT), interfractional and intrafractional imaging technologies are becoming increasingly critical to ensure safe and effective treatment delivery. Ultrasound (US)-based image guidance systems offer real-time, markerless, volumetric imaging with excellent soft tissue contrast, overcoming the limitations of traditional X-ray or computed tomography (CT)-based guidance for abdominal and pelvic cancer sites, such as the liver and prostate. Interfractional US guidance systems have been commercially adopted for patient positioning but suffer from systematic positioning errors induced by probe pressure. More recently, several research groups have introduced concepts for intrafractional US guidance systems leveraging robotic probe placement technology and real-time soft tissue tracking software. This paper reviews various commercial and research-level US guidance systems used in radiation therapy, with an emphasis on hardware and software technologies that enable the deployment of US imaging within the radiotherapy environment and workflow. Previously unpublished material on tissue tracking systems and robotic probe manipulators under development by our group is also included.

The non-trivial problem of Non-Line-of-Sight localization can be attacked with the aid of geometrical channel and mobility models. This approach results in high performance algorithms but requires a combination of different sources of information, like e.g. angles, delays and Doppler shifts. In this contribution we demonstrate the high performance of such methods even in poor scattering environments. On top of that, we compare the algorithms for different scenarios and highlight the superiority of algorithms designed for channels that change dynamically due to the movement of the mobile terminal.