Quantitative photoacoustic imaging method on basis of deep neural networks

Abstract

The invention relates to a quantitative photoacoustic imaging method on the basis of deep neural networks. The quantitative photoacoustic imaging method is characterized by comprising contents of constructing deep neural network frameworks, namely ResU-net; training the ResU-net by the aid of initial acoustic pressure images inputted at different wavelengths and corresponding quantitative images;carrying out quantitative photoacoustic imaging on multi-wavelength initial acoustic pressure images by the aid of trained ResU-net and outputting quantitative images. The quantitative photoacoustic imaging method has the advantages that residual learning mechanisms are used for the quantitative photoacoustic imaging deep neural networks, namely the ResU-net, accordingly, networks are easy to optimize and can reach considerable depths, and the quantitative photoacoustic imaging method is high in accuracy; furthermore, contraction paths and expansion paths are set for the deep neural networks,accordingly, comprehensive environment information of various resolution levels can be extracted from the inputted multi-wavelength initial acoustic pressure images by the ResU-net, and the high-resolution quantitative images can be ultimately outputted.

Description

technical field [0001] The invention relates to a photoacoustic imaging method, in particular to a quantitative photoacoustic imaging method based on a deep neural network. Background technique [0002] Photoacoustic (PA) imaging can achieve better spatial resolution and good specificity. Quantitative photoacoustic imaging (quantitative PA imaging, QPAI) can convert multispectral PA images into a series of precise images: such as concentration images of specific molecular markers for molecular imaging, assessment of tumor growth, metabolism, and response to various treatments. Resistance blood oxygen saturation chart, etc. [0003] The traditional QPAI method ignores the influence of wavelength on luminous flux, and uses linear fitting to estimate SO 2 (ie linear unmixing, linear unmixing) will bring serious errors. Most luminous flux correction methods rely on strong assumptions, such as optical properties have piecewise constant characteristics, scattering parameters ne...

AI Technical Summary

Problems solved by technology

[0003] The traditional QPAI method ignores the influence of wavelength on luminous flux, and uses linear fitting to estimate SO 2 (ie linear unmixing, linear unmixing) will bring serious errors

Most luminous flux correction methods rely on strong assumptions, such as optical properties have piecewise constant characteristics, scattering parameters need to be known in advance, background optical parameters are required to be uniform (and known), etc., or reconstruction calculations for large-scale high-resolution images too much

Images