Automatic Muscle Segmentation for the Diagnosis of Peripheral Artery Disease Using Multispectral Optoacoustic Tomography

Schillinger M, Schlereth M, Boraud G, Zahnd G, Dehner C, Li Y, Kempf J, Caranovic M, Haney B, Rother U, Breininger K (2025)

Publication Type: Conference contribution

Publication year: 2025

Journal

Progress in Biomedical Optics and Imaging - Proceedings of SPIE

Publisher: SPIE

Book Volume: 13319

Conference Proceedings Title: Progress in Biomedical Optics and Imaging - Proceedings of SPIE

Event location: San Francisco, CA, USA

ISBN: 9781510683860

DOI: 10.1117/12.3049067

Abstract

Multispectral optoacoustic tomography (MSOT) is a medical imaging modality capable of visualizing chromophore concentrations, such as oxygenated (HbO2) and deoxygenated hemoglobin (Hb), making it particularly useful for diagnosing blood-perfusion-related diseases like peripheral artery disease (PAD). Previous MSOT-based diagnostic studies involved experts manually selecting a region of interest (ROI) with a predefined shape in the target muscle to analyze blood oxygenation. This study aims to automate this process using a deep-learning-based approach. We present a pipeline that uses co-registered ultrasound images to automatically place an ROI necessary for PAD diagnosis in the MSOT image by training a deep-learning-based segmentation model to locate the target muscle. The resulting segmentation mask is overlaid onto the corresponding optoacoustic image to extract mean HbO2 and Hb intensities within the ROIs. We evaluated our pipeline on two separate PAD-related MSOT-datasets. When comparing blood oxygenation in the automatically generated ROIs on the MSOT images to identify PAD patients, we achieved areas under the ROC curve (AUCs) of 0.87 (vs. 0.85 for manually drawn ROIs) and 0.76 (vs. 0.77 for manually drawn ROIs) on the respective datasets. Our results indicate that optoacoustic ROI placement using deep-learning-based ultrasound segmentation is feasible and performs comparably to manually drawn ROIs by clinical experts. This approach could reduce the annotation effort in future MSOT studies and provide an ROI with greater physiological relevance compared to the arbitrary shapes used in related MSOT research. Future work could extend this approach to other body sites and diseases.

Authors with CRIS profile

Moritz Schillinger Department Artificial Intelligence in Biomedical Engineering (AIBE) Maja Schlereth Department Artificial Intelligence in Biomedical Engineering (AIBE) Yi Li Department of Vascular Surgery Julius Kempf Department of Vascular Surgery Briain Haney Department of Vascular Surgery Ulrich Rother Department of Vascular Surgery Katharina Breininger Department Artificial Intelligence in Biomedical Engineering (AIBE)

Involved external institutions

LMU Klinikum DE Germany (DE)

How to cite

APA:

Schillinger, M., Schlereth, M., Boraud, G., Zahnd, G., Dehner, C., Li, Y.,... Breininger, K. (2025). Automatic Muscle Segmentation for the Diagnosis of Peripheral Artery Disease Using Multispectral Optoacoustic Tomography. In Alexander A. Oraevsky, Lihong V. Wang (Eds.), Progress in Biomedical Optics and Imaging - Proceedings of SPIE. San Francisco, CA, USA: SPIE.

MLA:

Schillinger, Moritz, et al. "Automatic Muscle Segmentation for the Diagnosis of Peripheral Artery Disease Using Multispectral Optoacoustic Tomography." Proceedings of the Photons Plus Ultrasound: Imaging and Sensing 2025, San Francisco, CA, USA Ed. Alexander A. Oraevsky, Lihong V. Wang, SPIE, 2025.

BibTeX: Download