一种基于梯度下降的有限数据光声图像重建

Gradient descent-based photoacoustic image reconstruction with limited data

光声断层扫描(PAT)是一种新兴的基于光声效应的混合生物医学成像技术,具有高空间分辨率和高对比度。在实际的光声成像中,由于几何尺寸的限制或加速数据采集,探测器采集的光声数据通常是不完整的(有限检测视角和稀疏采样),这将导致重建图像模糊且存在伪影。为了解决有限数据PAT的问题,研究了一种基于梯度最小化的光声图像重建算法。该算法首先使用梯度下降方法获得初始解,然后将其投影到正向约束集上,最后采用二次平滑正则化项进一步提高重建质量。仿真实验结果表明,在有限视角和稀疏采样情况下,算法可以有效抑制图像噪声和伪影,具有较好的重建效果,与经典的时间反转(TR)、插值(ITR)和滤波反投影(FBP)重建算法相比,峰值信噪比(PSNR)分别提高约8.09、5.12和13.84 dB。

Photoacoustic tomography(PAT)is an emerging hybrid biomedical imaging technique based on photoacoustic effects with high spatial resolution and high contrast.In practical photoacoustic imaging,the data acquired by the detector is usually incomplete(limited detection view and sparse sampling)due to geometric limitations or accelerated data acquisition,which leads to blurred reconstructed images with artifacts.In order to solve the problem of limited data PAT,an optoacoustic image reconstruction algorithm based on gradient minimisation is studied.The algorithm first uses the gradient descent method to obtain an initial solution,then projects it onto the positive constraint set,and finally uses a quadratic smoothing regularization term to further improve the reconstruction quality.The simulation results show that the algorithm studied in this paper can effectively suppress the image noise and artifacts in the case of limited viewing angle and sparse sampling,and has a better reconstruction effect,and improves the peak signal-to-noise ratios(PSNRs)by about 8.09,5.12 and 13.84 dB,compared with those of the classical TR,ITR and FBP reconstruction algorithms,respectively.