Deep Learning Delivers Optoacoustic Images in Real Time

High-quality medical imaging from multispectral optoacoustic tomography (MSOT) could be used to diagnose and evaluate a range of diseases, including breast cancer, Duchenne muscular dystrophy, inflammatory bowel disease, and many more. However, the length of time currently required for MSOT to process high-quality images makes it impractical in clinical settings.

The MSOT scanner, developed by researchers from the Bioengineering Center and the Computational Health Center at Helmholtz Munich and the Technical University of Munich, collects sound waves generated when light is absorbed by tissue. It uses reconstruction algorithms to translate the sound waves into images, which are displayed on a monitor.

Some of the algorithms used by MSOT can reconstruct images quickly enough for the scanner to display in real time, but with low image quality. The more complex algorithms can produce high-quality images but require prohibitively long processing times.
DeepMB provides a deep learning framework for realizing high-quality optoacoustic imaging in real-time, enabling multispectral optoacoustic tomography (MSOT) to be used in clinical settings. Courtesy of iThera Medical GmbH/Guillaume Zahnd.

DeepMB provides a deep-learning framework for realizing high-quality optoacoustic imaging in real time, enabling multispectral optoacoustic tomography to be used in clinical settings. Courtesy of iThera Medical GmbH/Guillaume Zahnd.

Deep learning can enable fast reconstruction of optoacoustic images, but a lack of experimental ground-truth training data can lead to reduced image quality for in vivo scans. Model-based reconstruction can produce high-quality optoacoustic images, but it cannot be used for real-time imaging.

To provide high-quality, real-time optoacoustic imaging via MSOT, the researchers developed DeepMB, a deep-learning framework. DeepMB expresses model-based reconstruction with a deep neural network.

CASTECH INC - High Precision CNC Polished Aspherical Lenses

The researchers based their training strategy for DeepMB on optoacoustic signals synthesized from real-world images, paired with ground-truth optoacoustic images generated by model-based reconstruction. Using experimental data with DeepMB, they demonstrated accurate optoacoustic image reconstruction in 31 milliseconds per image.

The team further showed that DeepMB could reconstruct images approximately 1000 times faster than a state-of-the-art algorithm, with virtually no loss in image quality, based on qualitative and quantitative evaluations of a diverse data set of in vivo images.

According to the researchers, DeepMB can be used by clinicians to accurately reconstruct any scan acquired from a patient, regardless of the area of the body being scanned or the type of disease being evaluated.

Once deployed in the field, DeepMB will provide clinicians, for the first time, with direct access to optimal image quality from MSOT scans. Accurate, real-time image reconstruction with DeepMB will give medical personnel full access to the high-resolution, multispectral contrast of hand-held optoacoustic tomography, enabling this technology to be adopted into clinical routines.

The researchers believe that DeepMB represents a major step forward for optoacoustic imaging. High-quality, real-time optoacoustic imaging could have a positive impact on clinical studies and ultimately could help patients receive better care.

The core principles of DeepMB are adaptable and can be applied to other reconstruction methods in optoacoustic imaging, including other research efforts at Helmholtz Munich. The DeepMB framework could be applied to other imaging modalities, such as ultrasound, x-ray, or magnetic resonance imaging.

The research was led by professor Vasilis Ntziachristos and is jointly advanced by his research team at Helmholtz Munich and the Technical University of Munich and by his spinoff company, iThera Medical GmbH.

The research was published in Nature Machine Intelligence (www.doi.org/10.1038/s42256-023-00724-3).

Published: October 2023

Glossary

artificial intelligence: The ability of a machine to perform certain complex functions normally associated with human intelligence, such as judgment, pattern recognition, understanding, learning, planning, and problem solving.
deep learning: Deep learning is a subset of machine learning that involves the use of artificial neural networks to model and solve complex problems. The term "deep" in deep learning refers to the use of deep neural networks, which are neural networks with multiple layers (deep architectures). These networks, often called deep neural networks or deep neural architectures, have the ability to automatically learn hierarchical representations of data. Key concepts and components of deep learning include: ...
machine vision: Machine vision, also known as computer vision or computer sight, refers to the technology that enables machines, typically computers, to interpret and understand visual information from the world, much like the human visual system. It involves the development and application of algorithms and systems that allow machines to acquire, process, analyze, and make decisions based on visual data. Key aspects of machine vision include: Image acquisition: Machine vision systems use various...
multispectral imaging: Multispectral imaging is a technique that involves capturing and analyzing images at multiple discrete spectral bands within the electromagnetic spectrum. Unlike hyperspectral imaging, which acquires data across a continuous range of wavelengths, multispectral imaging is characterized by capturing information at several specific, predefined bands. This allows for the extraction of spectral signatures and information from different parts of the spectrum. Key aspects of multispectral imaging...
tomography: Technique that defocuses activity from surrounding planes by means of the relative motions at the point of interest.

Browse Cameras & Imaging, Lasers, Optical Components, Test & Measurement, and more.