Towards non-contact time-domain diffuse optical tomography

Ifechi Ejidike; Michael G. Tanner

Towards non-contact time-domain diffuse optical tomography

^*Corresponding author for this work

Research output: Contribution to conference › Poster › peer-review

Abstract

Near-infrared light has good contrast in scattering and absorption across different tissue types [1], diffuse optical tomography (DOT) leverages this by using computational techniques to reconstruct volumetric images of tissue absorption and scattering coefficients [2]. Rich data available in timedomain (TD) measurements enables high quality reconstructions but is victim to extremely long computation times, bottlenecked by finite-difference methods [3]. Non-contact methods [4] also suffer from this as the modelling amounts to a ray-tracing problem. Current DOT methods require specialised hardware to be worn by the patient [5] or the use of an intralipid to fill void space [6] which attenuates signal and can be uncomfortable for the patient. Non-contact DOT methods could realise a similar modality to CT and MRI which is familiar to both patients and clinicians. This enables multimodal approaches using structural priors [7], and compact single photon avalanche diodes (SPADs) could be easily integrated into such a system. Therefore, we present Diffuse Optics by Graphics Processing Unit Parallelisation (DOGPUP). A MATLAB toolkit which takes advantage of the parallel architecture of GPU hardware and the new CUDA support in MATLAB to enable fast simulation of non-contact TD-DOT detections, by decomposing the signal into a Fourier series [8] and using highly parallelisable Plücker coordinate raytriangle intersection tests [9]. We compare this to experimental data with a line scanning camera using a time-resolved CMOS SPAD [10].
[1]. Biomed Opt Express 11(3), pp. 1697-1706.
[2]. Commun Numer Methods Eng 25(6), pp. 711-732.
[3]. J. Opt. Soc. Am. A 37(12), pp. 1845-1856.
[4]. Physical review letters 91(10), pp. 103901-103904.
[5]. Nature Photonics 8(6), pp. 448-454.
[6]. Medical Physics 32(4), pp. 1128-1139.
[7]. J Xray Sci Technol 25(3), pp. 341-355, 2017.
[8]. J. Opt. Soc. Am. A 37(2), pp. 182-191.
[9]. Biomed. Opt. Express 1(1), pp. 165-175.
[10]. Opt. Express 30(15), pp. 27926-27937.

Original language	English
Publication status	Published - Sept 2024
Event	Photon 2024 - Swansea Arena, Swansea, United Kingdom Duration: 3 Sept 2024 → 6 Sept 2024 https://www.photon.org.uk/

Conference

Conference	Photon 2024
Country/Territory	United Kingdom
City	Swansea
Period	3/09/24 → 6/09/24
Internet address	https://www.photon.org.uk/

Cite this

@conference{cbdf6e76a7a14a2d918959dad5732d20,

title = "Towards non-contact time-domain diffuse optical tomography",

abstract = "Near-infrared light has good contrast in scattering and absorption across different tissue types [1], diffuse optical tomography (DOT) leverages this by using computational techniques to reconstruct volumetric images of tissue absorption and scattering coefficients [2]. Rich data available in timedomain (TD) measurements enables high quality reconstructions but is victim to extremely long computation times, bottlenecked by finite-difference methods [3]. Non-contact methods [4] also suffer from this as the modelling amounts to a ray-tracing problem. Current DOT methods require specialised hardware to be worn by the patient [5] or the use of an intralipid to fill void space [6] which attenuates signal and can be uncomfortable for the patient. Non-contact DOT methods could realise a similar modality to CT and MRI which is familiar to both patients and clinicians. This enables multimodal approaches using structural priors [7], and compact single photon avalanche diodes (SPADs) could be easily integrated into such a system. Therefore, we present Diffuse Optics by Graphics Processing Unit Parallelisation (DOGPUP). A MATLAB toolkit which takes advantage of the parallel architecture of GPU hardware and the new CUDA support in MATLAB to enable fast simulation of non-contact TD-DOT detections, by decomposing the signal into a Fourier series [8] and using highly parallelisable Pl{\"u}cker coordinate raytriangle intersection tests [9]. We compare this to experimental data with a line scanning camera using a time-resolved CMOS SPAD [10].[1]. Biomed Opt Express 11(3), pp. 1697-1706.[2]. Commun Numer Methods Eng 25(6), pp. 711-732.[3]. J. Opt. Soc. Am. A 37(12), pp. 1845-1856.[4]. Physical review letters 91(10), pp. 103901-103904.[5]. Nature Photonics 8(6), pp. 448-454.[6]. Medical Physics 32(4), pp. 1128-1139.[7]. J Xray Sci Technol 25(3), pp. 341-355, 2017.[8]. J. Opt. Soc. Am. A 37(2), pp. 182-191.[9]. Biomed. Opt. Express 1(1), pp. 165-175.[10]. Opt. Express 30(15), pp. 27926-27937.",

author = "Ifechi Ejidike and Tanner, {Michael G.}",

year = "2024",

month = sep,

language = "English",

note = "Photon 2024 ; Conference date: 03-09-2024 Through 06-09-2024",

url = "https://www.photon.org.uk/",

}

TY - CONF

T1 - Towards non-contact time-domain diffuse optical tomography

AU - Ejidike, Ifechi

AU - Tanner, Michael G.

PY - 2024/9

Y1 - 2024/9

N2 - Near-infrared light has good contrast in scattering and absorption across different tissue types [1], diffuse optical tomography (DOT) leverages this by using computational techniques to reconstruct volumetric images of tissue absorption and scattering coefficients [2]. Rich data available in timedomain (TD) measurements enables high quality reconstructions but is victim to extremely long computation times, bottlenecked by finite-difference methods [3]. Non-contact methods [4] also suffer from this as the modelling amounts to a ray-tracing problem. Current DOT methods require specialised hardware to be worn by the patient [5] or the use of an intralipid to fill void space [6] which attenuates signal and can be uncomfortable for the patient. Non-contact DOT methods could realise a similar modality to CT and MRI which is familiar to both patients and clinicians. This enables multimodal approaches using structural priors [7], and compact single photon avalanche diodes (SPADs) could be easily integrated into such a system. Therefore, we present Diffuse Optics by Graphics Processing Unit Parallelisation (DOGPUP). A MATLAB toolkit which takes advantage of the parallel architecture of GPU hardware and the new CUDA support in MATLAB to enable fast simulation of non-contact TD-DOT detections, by decomposing the signal into a Fourier series [8] and using highly parallelisable Plücker coordinate raytriangle intersection tests [9]. We compare this to experimental data with a line scanning camera using a time-resolved CMOS SPAD [10].[1]. Biomed Opt Express 11(3), pp. 1697-1706.[2]. Commun Numer Methods Eng 25(6), pp. 711-732.[3]. J. Opt. Soc. Am. A 37(12), pp. 1845-1856.[4]. Physical review letters 91(10), pp. 103901-103904.[5]. Nature Photonics 8(6), pp. 448-454.[6]. Medical Physics 32(4), pp. 1128-1139.[7]. J Xray Sci Technol 25(3), pp. 341-355, 2017.[8]. J. Opt. Soc. Am. A 37(2), pp. 182-191.[9]. Biomed. Opt. Express 1(1), pp. 165-175.[10]. Opt. Express 30(15), pp. 27926-27937.

AB - Near-infrared light has good contrast in scattering and absorption across different tissue types [1], diffuse optical tomography (DOT) leverages this by using computational techniques to reconstruct volumetric images of tissue absorption and scattering coefficients [2]. Rich data available in timedomain (TD) measurements enables high quality reconstructions but is victim to extremely long computation times, bottlenecked by finite-difference methods [3]. Non-contact methods [4] also suffer from this as the modelling amounts to a ray-tracing problem. Current DOT methods require specialised hardware to be worn by the patient [5] or the use of an intralipid to fill void space [6] which attenuates signal and can be uncomfortable for the patient. Non-contact DOT methods could realise a similar modality to CT and MRI which is familiar to both patients and clinicians. This enables multimodal approaches using structural priors [7], and compact single photon avalanche diodes (SPADs) could be easily integrated into such a system. Therefore, we present Diffuse Optics by Graphics Processing Unit Parallelisation (DOGPUP). A MATLAB toolkit which takes advantage of the parallel architecture of GPU hardware and the new CUDA support in MATLAB to enable fast simulation of non-contact TD-DOT detections, by decomposing the signal into a Fourier series [8] and using highly parallelisable Plücker coordinate raytriangle intersection tests [9]. We compare this to experimental data with a line scanning camera using a time-resolved CMOS SPAD [10].[1]. Biomed Opt Express 11(3), pp. 1697-1706.[2]. Commun Numer Methods Eng 25(6), pp. 711-732.[3]. J. Opt. Soc. Am. A 37(12), pp. 1845-1856.[4]. Physical review letters 91(10), pp. 103901-103904.[5]. Nature Photonics 8(6), pp. 448-454.[6]. Medical Physics 32(4), pp. 1128-1139.[7]. J Xray Sci Technol 25(3), pp. 341-355, 2017.[8]. J. Opt. Soc. Am. A 37(2), pp. 182-191.[9]. Biomed. Opt. Express 1(1), pp. 165-175.[10]. Opt. Express 30(15), pp. 27926-27937.

M3 - Poster

T2 - Photon 2024

Y2 - 3 September 2024 through 6 September 2024

ER -

Towards non-contact time-domain diffuse optical tomography

Abstract

Conference

Fingerprint

Cite this