Signal domain adaptation network for limited-view optoacoustic tomography

Area of Science:

• Biomedical Optics
• Medical Imaging
• Acoustics

Background:

• Optoacoustic (OA) imaging relies on detecting ultrasound (US) waves from laser-excited tissues, with image quality dependent on US detector coverage.
• Limited angular coverage from transducer arrays causes artifacts in OA images, especially with hybrid OA/US systems and handheld probes.
• Current deep learning methods often struggle with experimental data due to reliance on simulated datasets and lack of ground truth.

Purpose of the Study:

• To develop a deep learning method that improves optoacoustic image reconstruction from limited-view data.
• To address the challenge of domain gap between simulated and experimental OA signals.
• To eliminate the need for ground truth data from full tomographic acquisitions in clinical settings.

Main Methods:

• Proposed a signal domain adaptation network (SDAN) comprising a domain adaptation network and a sides prediction network.
• SDAN reduces the domain gap between simulated and experimental OA signals.
• Complemented missing signals in limited-view OA datasets acquired from a human forearm using a handheld linear transducer array.

Main Results:

• The proposed SDAN method demonstrated improved performance in reducing limited-view artifacts in optoacoustic images.
• The approach successfully minimized artifacts without requiring ground truth signals from full tomographic acquisitions.
• Effective signal domain adaptation was achieved, enhancing OA image fidelity.

Conclusions:

• The signal domain adaptation network (SDAN) offers a promising solution for improving optoacoustic imaging in scenarios with experimental constraints.
• This method enhances the practical applicability of deep learning in optoacoustic imaging, particularly with handheld devices.
• SDAN overcomes limitations of previous deep learning approaches by effectively handling the domain gap between simulated and real-world data.