Header

UZH-Logo

Maintenance Infos

Efficient segmentation of multi-modal optoacoustic and ultrasound images using convolutional neural networks


Lafci, Berkan; Mercep, Elena; Morscher, Stefan; Deán-Ben, Xosé Luís; Razansky, Daniel (2020). Efficient segmentation of multi-modal optoacoustic and ultrasound images using convolutional neural networks. In: Photons Plus Ultrasound: Imaging and Sensing 2020, San Francisco, 1 February 2020 - 6 February 2020.

Abstract

Multispectral optoacoustic tomography (MSOT) offers the unique capability to map the distribution of spectrally distinctive endogenous and exogenous substances in heterogeneous biological tissues by exciting the sample at various wavelengths and detecting the optoacoustically-induced ultrasound waves. This powerful functional and molecular imaging capability can greatly benefit from hybridization with pulse-echo ultrasound (US), which provides additional information on tissue anatomy and blood flow. However, speed of sound variations and acoustic mismatches in the imaged object generally lead to errors in the coregistration of compounded images and loss of spatial resolution in both imaging modalities. The spatially- and wavelength-dependent light fluence attenuation further limits the quantitative capabilities of MSOT. Proper segmentation of different regions and assignment of corresponding acoustic and optical properties turns then essential for maximizing the performance of hybrid optoacoustic and ultrasound (OPUS) imaging. Particularly, accurate segmentation of the boundary of the sample can significantly improve the images rendered. Herein, we propose an automatic segmentation method based on a convolutional neural network (CNN) for segmenting the mouse boundary in a pre-clinical OPUS system. The experimental performance of the method, as characterized with the Dice coefficient metric between the network output and the ground truth (manually segmented) images, is shown to be superior than that of a state-of-the-art active contour segmentation method in a series of two-dimensional (cross-sectional) OPUS images of the mouse brain, liver and kidney regions.

Abstract

Multispectral optoacoustic tomography (MSOT) offers the unique capability to map the distribution of spectrally distinctive endogenous and exogenous substances in heterogeneous biological tissues by exciting the sample at various wavelengths and detecting the optoacoustically-induced ultrasound waves. This powerful functional and molecular imaging capability can greatly benefit from hybridization with pulse-echo ultrasound (US), which provides additional information on tissue anatomy and blood flow. However, speed of sound variations and acoustic mismatches in the imaged object generally lead to errors in the coregistration of compounded images and loss of spatial resolution in both imaging modalities. The spatially- and wavelength-dependent light fluence attenuation further limits the quantitative capabilities of MSOT. Proper segmentation of different regions and assignment of corresponding acoustic and optical properties turns then essential for maximizing the performance of hybrid optoacoustic and ultrasound (OPUS) imaging. Particularly, accurate segmentation of the boundary of the sample can significantly improve the images rendered. Herein, we propose an automatic segmentation method based on a convolutional neural network (CNN) for segmenting the mouse boundary in a pre-clinical OPUS system. The experimental performance of the method, as characterized with the Dice coefficient metric between the network output and the ground truth (manually segmented) images, is shown to be superior than that of a state-of-the-art active contour segmentation method in a series of two-dimensional (cross-sectional) OPUS images of the mouse brain, liver and kidney regions.

Statistics

Citations

Altmetrics

Downloads

0 downloads since deposited on 01 Feb 2021
0 downloads since 12 months

Additional indexing

Item Type:Conference or Workshop Item (Paper), refereed, original work
Communities & Collections:04 Faculty of Medicine > Institute of Pharmacology and Toxicology
07 Faculty of Science > Institute of Pharmacology and Toxicology

04 Faculty of Medicine > Institute of Biomedical Engineering
Dewey Decimal Classification:170 Ethics
610 Medicine & health
Language:English
Event End Date:6 February 2020
Deposited On:01 Feb 2021 13:35
Last Modified:23 Feb 2021 20:53
Publisher:Spie
ISBN:9781510632431
OA Status:Green
Publisher DOI:https://doi.org/10.1117/12.2543970

Download

Green Open Access

Download PDF  'Efficient segmentation of multi-modal optoacoustic and ultrasound images using convolutional neural networks'.
Preview
Content: Published Version
Filetype: PDF
Size: 1MB
View at publisher