Recurrent and convolutional neural networks for sequential multispectral optoacoustic tomography (MSOT) imaging.

Area of Science:

• Biomedical Imaging
• Medical Technology
• Artificial Intelligence in Medicine

Background:

• Multispectral optoacoustic tomography (MSOT) offers detailed anatomical and physiological insights into biological samples.
• Acquiring high-resolution volumetric MSOT data is currently a time-intensive process.

Purpose of the Study:

• To develop a deep learning model for accelerating volumetric MSOT image acquisition.
• To reduce the time required for obtaining high through-plane resolution MSOT data.

Main Methods:

• A hybrid recurrent and convolutional neural network model was designed for sequential image generation.
• The model processes two sparsely sampled MSOT images (0.6 mm step size) to generate densely sampled images (0.1 mm step size).
• Indocyanine green-conjugated nanoworms (NWs-ICG) were utilized as the contrast agent in a multimodal imaging system.

Main Results:

• The deep learning model successfully generated five intermediate cross-sectional images between two input images.
• This approach reduced MSOT acquisition time by approximately 71% compared to conventional methods.
• The system integrates MSOT, ultrasound, and optoacoustic imaging in a single scan.

Conclusions:

• The proposed deep learning model significantly accelerates volumetric MSOT image acquisition.
• This method enhances the efficiency of analyzing biological samples using MSOT.
• The hybrid neural network approach offers a promising solution for time-consuming volumetric imaging challenges.