BPU-Net: An efficient end-to-end deep-learning diffuse optical tomography model for structured scan data

Ben Fry; Rickson C. Mesquita; Hamid Dehghani; Robin Dale

doi:10.1117/12.3098433

BPU-Net: An efficient end-to-end deep-learning diffuse optical tomography model for structured scan data

Ben Fry^*
, Rickson C. Mesquita
, Hamid Dehghani
, Robin Dale

^*Corresponding author for this work

Computer Science

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

Current Deep Learning Diffuse Optical Tomography (DL-DOT) architectures do not exploit dataset structures, causing learning inefficiencies. This work proposes a network design that considers integration of spatial structures, which caused a 44% training time reduction, increased performance by up to 43% across a series of metrics and at least a 40% reduction in training epochs regardless of dataset size.

Original language	English
Title of host publication	Diffuse Optical Spectroscopy and Imaging X
Editors	Davide Contini, Yoko Hoshi, Thomas D. O'Sullivan
Publisher	SPIE
Number of pages	3
ISBN (Electronic)	9781510698079
DOIs	https://doi.org/10.1117/12.3098433
Publication status	Published - 18 Dec 2025
Event	10th Diffuse Optical Spectroscopy and Imaging - Munich, Germany Duration: 22 Jun 2025 → 26 Jun 2025

Publication series

Name	Progress in Biomedical Optics and Imaging
Publisher	SPIE
Volume	13935
ISSN (Print)	1605-7422
ISSN (Electronic)	2410-9045

Conference

Conference	10th Diffuse Optical Spectroscopy and Imaging
Country/Territory	Germany
City	Munich
Period	22/06/25 → 26/06/25

Bibliographical note

Publisher Copyright:
© 2026 SPIE.

ASJC Scopus subject areas

Electronic, Optical and Magnetic Materials
Atomic and Molecular Physics, and Optics
Biomaterials
Radiology Nuclear Medicine and imaging

Access to Document

10.1117/12.3098433

Cite this

Fry, B., Mesquita, R. C., Dehghani, H., & Dale, R. (2025). BPU-Net: An efficient end-to-end deep-learning diffuse optical tomography model for structured scan data. In D. Contini, Y. Hoshi, & T. D. O'Sullivan (Eds.), Diffuse Optical Spectroscopy and Imaging X Article 139351O (Progress in Biomedical Optics and Imaging; Vol. 13935). SPIE. https://doi.org/10.1117/12.3098433

@inproceedings{fb4e6d91640d4aed81b6a6c9399443a0,

title = "BPU-Net: An efficient end-to-end deep-learning diffuse optical tomography model for structured scan data",

abstract = "Current Deep Learning Diffuse Optical Tomography (DL-DOT) architectures do not exploit dataset structures, causing learning inefficiencies. This work proposes a network design that considers integration of spatial structures, which caused a 44\% training time reduction, increased performance by up to 43\% across a series of metrics and at least a 40\% reduction in training epochs regardless of dataset size.",

author = "Ben Fry and Mesquita, \{Rickson C.\} and Hamid Dehghani and Robin Dale",

note = "Publisher Copyright: {\textcopyright} 2026 SPIE.; 10th Diffuse Optical Spectroscopy and Imaging ; Conference date: 22-06-2025 Through 26-06-2025",

year = "2025",

month = dec,

day = "18",

doi = "10.1117/12.3098433",

language = "English",

series = "Progress in Biomedical Optics and Imaging",

publisher = "SPIE",

editor = "Davide Contini and Yoko Hoshi and O'Sullivan, \{Thomas D.\}",

booktitle = "Diffuse Optical Spectroscopy and Imaging X",

}

Fry, B, Mesquita, RC, Dehghani, H & Dale, R 2025, BPU-Net: An efficient end-to-end deep-learning diffuse optical tomography model for structured scan data. in D Contini, Y Hoshi & TD O'Sullivan (eds), Diffuse Optical Spectroscopy and Imaging X., 139351O, Progress in Biomedical Optics and Imaging, vol. 13935, SPIE, 10th Diffuse Optical Spectroscopy and Imaging, Munich, Germany, 22/06/25. https://doi.org/10.1117/12.3098433

BPU-Net: An efficient end-to-end deep-learning diffuse optical tomography model for structured scan data. / Fry, Ben; Mesquita, Rickson C.; Dehghani, Hamid et al.
Diffuse Optical Spectroscopy and Imaging X. ed. / Davide Contini; Yoko Hoshi; Thomas D. O'Sullivan. SPIE, 2025. 139351O (Progress in Biomedical Optics and Imaging; Vol. 13935).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - BPU-Net

T2 - 10th Diffuse Optical Spectroscopy and Imaging

AU - Fry, Ben

AU - Mesquita, Rickson C.

AU - Dehghani, Hamid

AU - Dale, Robin

PY - 2025/12/18

Y1 - 2025/12/18

N2 - Current Deep Learning Diffuse Optical Tomography (DL-DOT) architectures do not exploit dataset structures, causing learning inefficiencies. This work proposes a network design that considers integration of spatial structures, which caused a 44% training time reduction, increased performance by up to 43% across a series of metrics and at least a 40% reduction in training epochs regardless of dataset size.

AB - Current Deep Learning Diffuse Optical Tomography (DL-DOT) architectures do not exploit dataset structures, causing learning inefficiencies. This work proposes a network design that considers integration of spatial structures, which caused a 44% training time reduction, increased performance by up to 43% across a series of metrics and at least a 40% reduction in training epochs regardless of dataset size.

UR - https://www.scopus.com/pages/publications/105026752168

U2 - 10.1117/12.3098433

DO - 10.1117/12.3098433

M3 - Conference contribution

AN - SCOPUS:105026752168

T3 - Progress in Biomedical Optics and Imaging

BT - Diffuse Optical Spectroscopy and Imaging X

A2 - Contini, Davide

A2 - Hoshi, Yoko

A2 - O'Sullivan, Thomas D.

PB - SPIE

Y2 - 22 June 2025 through 26 June 2025

ER -

BPU-Net: An efficient end-to-end deep-learning diffuse optical tomography model for structured scan data

Abstract

Publication series

Conference

Bibliographical note

ASJC Scopus subject areas

Access to Document

Fingerprint

Cite this