AIM：To determine the Bruch＇s membrane opening-minimum rim width（BMO-MRW） tipping point where corresponding visual field（VF） damages become detectable.METHODS：A total of 85 normal subjects and 83 glaucoma patie...AIM：To determine the Bruch＇s membrane opening-minimum rim width（BMO-MRW） tipping point where corresponding visual field（VF） damages become detectable.METHODS：A total of 85 normal subjects and 83 glaucoma patients（one eye per participant） were recruited for the study.All of the patients had VF examinations and spectral-domain optical coherence tomography to measure the BMO-MRW.Total deviation values for 52 VF points were allocated to the corresponding sector according to the Garway-Heath distribution map.To evaluate the relationship between VF loss and BMOMRW measurements,a ＂broken-stick＂ statistical model was used.The tipping point where the VF values started to sharply decrease as a function of BMO-MRW measurements was estimated and the slopes above and below this tipping point were compared.RESULTS：A 25.9% global BMO-MRW loss from normative value was required for the VF loss to be detectable.Sectorally,substantial BMO-MRW thinning in inferotemporal sector（33.1%） and relatively less BMO-MRW thinning in the superotemporal sector（8.9%） were necessary for the detection of the VF loss.Beyond the tipping point,the slopes were close to zero throughout all of the sectors and the VF loss was unrelated to the BMO-MRW loss.The VF loss was related to the BMO-MRW loss below the tipping point.The difference between the two slopes was statistically significant（P≤0.002）.CONCLUSION：Substantial BMO-MRW loss appears to be necessary for VF loss to be detectable in patients with open angle glaucoma with standard achromatic perimetry.展开更多
针对破壳鸡蛋(破口蛋和裂纹蛋)缺陷差异性大,在线检测要求实时,以及人工检测依靠主观经验且检测速度慢、检测精度不高等问题,该研究提出一种基于改进的YOLOv7(You Only Look Once v7)模型的破壳鸡蛋在线实时检测系统。以YOLOv7网络为基...针对破壳鸡蛋(破口蛋和裂纹蛋)缺陷差异性大,在线检测要求实时,以及人工检测依靠主观经验且检测速度慢、检测精度不高等问题,该研究提出一种基于改进的YOLOv7(You Only Look Once v7)模型的破壳鸡蛋在线实时检测系统。以YOLOv7网络为基础,将YOLOv7网络的损失函数CIoU(complete-IoU)替换为WIoUv2(wise-IoU),在骨干网络(backbone)中嵌入坐标注意力模块(coordinate attention,CA)和添加可变形卷积DCNv2(deformable convnet)模块,同时将YOLOv7网络中的检测头(IDetect)替换为具有隐式知识学习的解耦检测头(IDetect_Decoupled)模块。在PC端的试验结果表明,改进后的模型在测试集上平均精度均值(mean average precision,m AP)为94.0%,单张图片检测时间为13.1 ms,与模型改进之前相比,其mAP提高了2.9个百分点,检测时间仅延长1.0 ms;改进后模型的参数量为3.64×10^(7),较原始模型降低了2.1%。最后通过格式转换并利用ONNXRun time深度学习框架把模型部署至设备端,在ONNXRuntime推理框架下进行在线检测验证。试验结果表明:该算法相较原始YOLOv7误检率降低了3.8个百分点,漏检率不变,并且在线检测平均帧率约为54帧/s,满足在线实时性检测需求。该研究可为破壳鸡蛋在线检测研究提供技术参考。展开更多
文摘AIM：To determine the Bruch＇s membrane opening-minimum rim width（BMO-MRW） tipping point where corresponding visual field（VF） damages become detectable.METHODS：A total of 85 normal subjects and 83 glaucoma patients（one eye per participant） were recruited for the study.All of the patients had VF examinations and spectral-domain optical coherence tomography to measure the BMO-MRW.Total deviation values for 52 VF points were allocated to the corresponding sector according to the Garway-Heath distribution map.To evaluate the relationship between VF loss and BMOMRW measurements,a ＂broken-stick＂ statistical model was used.The tipping point where the VF values started to sharply decrease as a function of BMO-MRW measurements was estimated and the slopes above and below this tipping point were compared.RESULTS：A 25.9% global BMO-MRW loss from normative value was required for the VF loss to be detectable.Sectorally,substantial BMO-MRW thinning in inferotemporal sector（33.1%） and relatively less BMO-MRW thinning in the superotemporal sector（8.9%） were necessary for the detection of the VF loss.Beyond the tipping point,the slopes were close to zero throughout all of the sectors and the VF loss was unrelated to the BMO-MRW loss.The VF loss was related to the BMO-MRW loss below the tipping point.The difference between the two slopes was statistically significant（P≤0.002）.CONCLUSION：Substantial BMO-MRW loss appears to be necessary for VF loss to be detectable in patients with open angle glaucoma with standard achromatic perimetry.
文摘针对破壳鸡蛋(破口蛋和裂纹蛋)缺陷差异性大,在线检测要求实时,以及人工检测依靠主观经验且检测速度慢、检测精度不高等问题,该研究提出一种基于改进的YOLOv7(You Only Look Once v7)模型的破壳鸡蛋在线实时检测系统。以YOLOv7网络为基础,将YOLOv7网络的损失函数CIoU(complete-IoU)替换为WIoUv2(wise-IoU),在骨干网络(backbone)中嵌入坐标注意力模块(coordinate attention,CA)和添加可变形卷积DCNv2(deformable convnet)模块,同时将YOLOv7网络中的检测头(IDetect)替换为具有隐式知识学习的解耦检测头(IDetect_Decoupled)模块。在PC端的试验结果表明,改进后的模型在测试集上平均精度均值(mean average precision,m AP)为94.0%,单张图片检测时间为13.1 ms,与模型改进之前相比,其mAP提高了2.9个百分点,检测时间仅延长1.0 ms;改进后模型的参数量为3.64×10^(7),较原始模型降低了2.1%。最后通过格式转换并利用ONNXRun time深度学习框架把模型部署至设备端,在ONNXRuntime推理框架下进行在线检测验证。试验结果表明:该算法相较原始YOLOv7误检率降低了3.8个百分点,漏检率不变,并且在线检测平均帧率约为54帧/s,满足在线实时性检测需求。该研究可为破壳鸡蛋在线检测研究提供技术参考。
