24-29 March 2024
BHSS, Academia Sinica
Asia/Taipei timezone

3D Ultrasound Computer Tomography - a data & computing intensive approach at multimodal ultrasound imaging

Mar 27, 2024, 4:30 PM
30m
Auditorium (BHSS, Academia Sinica)

Auditorium

BHSS, Academia Sinica

Oral Presentation Track 2: Health & Life Sciences (including Pandemic Preparedness Applications) Health & Life Science Applications

Speaker

Michael zapf

Description

Background & Motivation

3D Ultrasound Computer Tomography (3D USCT) is developed at the
Karlsruhe Institute of Technology for early breast cancer detection.
Unlike conventional ultrasound sonography methods with manual guided
ultrasound (US) probes, the patient is placed on a patient bed with a
stable and reproducible measurement configuration. A reproducible stable
measurement configuration is achieved by surrounding the patient's breast
with many spherical placed ultrasound transducers -- 2304 individual US
transducer elements. This allows screening and diagnosis in a
reproducible way -- long term, longitudinal tracking of a patient's
status of it's breast health is possible with a non-ionizing, 3D and
easy to delivery imaging modality is now becoming thinkable.

Method

Despite the promising design and vision of the project, early
pre-studies in hospitals (Hospital Jena, Hospital Mannheim) indicated
that there are still some technical challenges to be tackled to realise
the full potential of the method - the enormous computational burden of
the various imaging approaches is hindering the scientific progress and
the method.

More specifically, on challenge is the large amount of data (40 to
80 GB per measurement of pressure of time signals, so called a-scans)
combined with large three dimensional imaging domain, the
Region-of-interest, of 20x20x20cm\^3. With the desired resolution
of \~0.2mm this leads to an image volume of 1GVoxel in need of being
computed. In case of reflectivity image reconstruction with SAFT
(Synthetic Aperture Focussing Technique) both combined lead to 0.112
Terabyte (double floatingpoint data type) in write and read accesses in
the computing process of the image.

While the SAFT algorithm is parallelised in GPU, the other modalities
imaging methods having their challenges in the dimensionality and
non-trivial partitioning schemes - the transmission based tomography is
formulated as optimization problem, which is challenging due to the
inherent three dimensional nature oft he USCT's aperture.

Even more demanding approaches of promising full wave inversion schemes,
inspired also from prior work in the field of geo-science, which struggle
with high-frequency nature of the data provided by the USCT device and
method.

Next Steps

HPC and grid computing should be offering the infrastructure and
interfaces to tackle the multifold computational partitioning and big data
challenges of USCT. The USCT project tries now to enable fellow scientist
and associated communities and kick off collaborations. We are committed
to open science, open access and open data: example data sets and access
code are provided available under liberal licenses on github and an
webserver.

A Matlab script with some reference imaging and visualization code:

"3D-USCT-III-access-script" KIT-3DUSCT/3D-USCT-III-access-script
(github.com)

The following datasets are provided:

1: Gelatine phantom with four inclusions made from PVC (spheres of
different size 8 mm to 22 mm)

2: Empty measurement with the same acquisition parameters as the
gelatin phantom.

3D KIT USCT -- USCT data exchange and
collaboration

Keywords: HPC, Bioinformatics, Medical imaging, open science, open data

Primary author

Co-authors

Presentation materials