Clinical Application of Scatter Correction with Artificial Neural Network in Myocardial and Brain SPECT

本文提出了一种新的基于人工神经网络（ANN）的散射校正算法，并在双同位素SPECT中进行了验证，用于分离Tc-99m和I-123的主光子。最初，我们安排了一个人工神经网络，其中一个输入层包含10个单元，需要10个能量窗口进行数据采集。该方法在模拟实验中SPECT定量误差在4%以内。然而，信噪比并不令人满意，而且目前的伽马相机系统无法实现能量设置。为了克服上述问题，设计了一种包含一个输入层和3个单元的更新神经网络，用于3个能量窗口的获取。幻影实验得到的图像噪声较低，SPECT定量误差在5%以内。采用脑SPECT和心肌SPECT对人工神经网络进行临床试验，同时评价静息与应激脑灌注、心肌灌注与脂肪酸代谢、心肌灌注与心脏交感神经功能。将锝-99m和碘-123图像清晰分离，得到具有临床可接受噪声的图像。这种方法使我们只需一次采集就可以获得包含两种不同类型信息的图像。

We have developed a novel scatter correction algorithm with an artificial neural network(ANN) and it was validated in dual isotope SPECT for separating the primary photons of Tc-99m from those of I-123.Initially, we arranged an ANN with one input layer comprising 10 units that requires 10 energy windows for data acquisition. This method resulted in errors within 4% in SPECT quantification in phantom experiments. However, the signal-to-noise ratios were not satisfactory, and the energy setting is not achievable in current gamma camera systems.A renewed ANN containing one input layer with 3 units for 3 energy window acquisition was designed to overcome the above problems. Phantom experiments yielded images with reduced noise, and errors of SPECT quantification within 5%.Clinical trials of the ANN were conducted in brain and myocardial SPECT for the simultaneous evaluation of rest and stress brain perfusion, myocardial perfusion and fatty acid metabolism, and myocardial perfusion and cardiac sympathetic nerve function. Technetium-99m and I-123 images were clearly separated and images with clinically acceptable noises were obtained.This method enables us to obtain images including two different pieces of information of various kinds by just one-time acquisition.