At SPIE Medical Imaging 2018 on 14th February a second panel discussion took place. The discussion was moderated by Nicole Ruiter (Karlsruhe Institute of Technology, Germany). As panelists we had the honor to welcome

  • Mark Anastasio, Washington University in St. Louis
  • Christian Böhm, Swiss Federal Institute of Technology Zurich (ETH)
  • Neb Duric, Delphinus Medical Technologies, Inc.
  • Hartmut Gemmeke, Karlsruhe Institute of Technology
  • Ken-ichi Kawabata, Hitachi, Ltd.
  • Roberto Lavarello, Pontifical Catholic University of Peru

The panelists and the audience discussed possible standardized formats for USCT data, well defined phantom and simulated data for inclusion in the USCT reference database and best options for a future challenge. A detailed summary of the discussion is given below.

Standardized formats

A standardized format for USCT raw data and metadata describing the data and system parameters was generally of interest for the panelists and the audience. Adapting the Adaptable Seismic Data Format (ASDF) based on the Hierarchical Data Format (HDF5) as used by the seismic community was found to be a candidate for adaption. Alternatively it was proposed to store the raw data in the Digital Imaging and Communications in Medicine (DICOM) format as it is widely used for medical images. The quite large raw datasets of USCT systems were also considered to be in need for a good tool for compression, which might be delivered by tools supporting the standardized formats.

Which metadata should be provided was also discussed. E.g. should an empty measurement (i.e. calibration data measuring only water) be included? It was also mentioned that the essential metadata to enable any reconstruction method is not known yet. It would be beneficial when a task group would form to prepare a first proposal for a USCT format, which could then be adopted and extended by the community. Partnering communities could be involved in this task group.

Well defined phantoms

In last year’s panel discussion all participants and panelists agreed that a lasting standard phantom with well-known ultrasound parameters would be beneficial. It could be imaged with different systems to compare systems and reconstruction methods. Yet, the currently used phantoms usually lack either well-known parameters or longlivety or both. We discussed the new development of a well characterized oil-gel-based phantom (see paper 10580-18), the possibility to acquire a defined phantom by one of the vendors or use a PVC based phantom. We agreed to start a webpage collecting phantoms, there sources or recipes and experiences of using these phantoms with USCT systems.

Simulated data

The participants and panelists agreed to include simulated raw data and numerical phantoms in the USCT reference database. The inclusion of typical pathologies like tumors or cysts would also be valuable. The Optical and Acoustical Breast Phantom Database (OA-Breast) and the  anthropomorphic software breast phantom of UPenn were mentioned as good examples for numerical phantoms.

Different simulation tools were discussed:  Salvus, k-wave and others.  There was a consensus that data generated with different simulation methods should be available so that no “inverse crime” will happen during iterative reconstructions also using simulations. A hierarchy of easy to hard datasets for reconstruction should be made available, e.g. including noise, variance in tissue parameters, increasingly adding realistic experimental limitations, etc. The added value of simulated data could also be that very detailed simulations requiring a lot of computational time would be made available to the community.

Future challenge

Most panelists and participants would favor the next challenge to be based on simulated data as then the ground truth would be known and low frequencies and simple system parameters could be employed to also enable computational expensive reconstruction algorithms. The data set should include realistic noise and experimental limitations, i.e. limited opening angle of transducers etc., to be challenging. Having more than one dataset with different levels of realism would be beneficial.

Further datasets and systems are very welcome and can be added to the USCT reference database.

Mostly reconstructions of speed of sound or attenuation distributions would be preferred as this seems to be currently the most active field of research. The quality measures employed to rank the contributions of the challenge have to be agreed on when the topic of the challenge is determined.

The challenge should be made public with enough time for the participants to work on the data, i.e. more than six month in advance. Partnering communities should also be notified and invited to participate. The next challenge shall be held at the next MUST workshop.
We thank all panelists and participants in the discussion for their contributions!