基于深度学习的SPECT图像左心室心肌感兴趣区自动识别技术

Automatic Recognition on Region of Interest for Left Ventricular Myocardium in SPECT Images Based on Deep Learning

目的设计并验证基于深度学习的单光子发射计算机断层成像仪(Single-Photon Emission Computed Tomography,SPECT)图像左心室心肌感兴趣区(Region of Interest,ROI)自动识别技术。方法设计一种基于神经网络的深度学习和全自动SPECT图像左心室心肌ROI目标检测的算法,使用30幅真实心肌灌注SPECT断层重建数据进行评估。结果本文所提出的基于深度学习的心肌自动识别技术增强了横断面切片心肌组织和周围本底的对比度,可正确识别左心室三维心肌ROI,与人工判断结果相符。结论本文设计的基于深度学习的SPECT图像左心室心肌ROI目标检测全自动算法可有效识别心肌三维ROI,可为核医学SPECT心肌定量分析研究奠定理论基础。

Objective To design and verify a deep learning-based single-photon emission computed tomography(SPECT)imaging technique for left ventricle region of interest(ROI).Methods A deep learning algorithm based on neural network and automatic SPECT image was designed to detect the left ventricular ROI.30 real myocardial perfusion SPECT tomography reconstruction data were used for evaluation.Results The automatic myocardial recognition technology based on deep learning proposed in this paper enhanced the contrast between the myocardial tissue and the surrounding background in the cross-section slice,and accurately identifies the 3D left ventricular ROI,which could consistent with the results of artificial judgment.Conclusion The fully automatic object-detection algorithm bases on deep learning described in this paper can automatically and effectively identify the 3D region of interest of myocardium in SPECT images,which also lays a solid foundation for the quantitative analysis of myocardium in SPECT nuclear medicine images.