基于深度学习的SPECT图像关节炎病灶分割

Arthritic Lesion Segmentation of SPECT Images Based on Deep Learning

SPECT骨显像是辅助医生诊断疾病的重要手段.医生依靠手工勾画提取病灶区域的方法效率低且具有一定的主观性.针对这一问题,提出R_U-Net网络模型分割关节炎病灶,不仅能节省医生的诊断时间,还能为患者争取最佳治疗时间.为了检测模型分割效果,使用测试集中含有关节炎病灶的图像分割测试,与Mask R-CNN网络和原UNet网络相比,R_U-Net网络对关节部位病灶分割结果有所提升,其MPA达到77.6%,MIoU指标达到75.4%.实验结果表明,基于U-Net改进的R_U-Net网络对于SPECT图像关节炎病灶分割效果更好.

SPECT bone imaging is an important tool to assist physicians in diagnosing SPECT diseases.In order to accurately diagnose the patient’s condition,doctors mainly rely on the manual delineation method to extract the lesion area,which is inefficient and subjective to a certain extent.To solve this problem,R-U-Net network model is proposed to segment the lesion of arthritis,which can not only save the doctor’s diagnosis time,but also gain the best treatment time for the patients.Firstly,the knowledge of medical image segmentation is introduced.Then the preprocessing of SPECT data is introduced.Finally,based on the Mask R-CNN network,U-Net network and R-U-Net network,the osteoarthritis lesions in bone imaging are segmented.In order to test the segmentation effect of the model,the image segmentation test with arthritis lesion in the test set was used.Compared with the Mask RCNN network and the original U-Net network,R-U-Net network improved the segmentation result of the lesion at the joint site,with its MPA and MIoU indexes reaching 77.6%and 75.4%respectively.The experimental results showed that R-U-Net network had better effect on osteoarthritis lesion segmentation of SPECT images.