目的探讨一种基于深度学习算法的原发性闭角型青光眼(PACG)睫状体前位自动评估的超声生物显微镜(UBM)图像分析系统的临床应用价值。
方法采用诊断试验研究方法，收集自2022年8月至2023年12月于武汉大学人民医院眼科中心进行UBM检查的PACG患者378例726眼的2 132张UBM图像，将数据集分为训练集1 599张图像和测试集533张图像，采用深度学习算法构建模型。选取于黄石爱尔眼科医院就诊的PACG患者69例101眼的334张UBM图像进行外部测试。另选取110张UBM图像作为独立数据集进行人机比赛，以比较睫状体前位评估系统与3名高年资眼科医师的准确度及速度。此外，选取8名低年资医师对独立于数据集的110张UBM图像分别在模型辅助前后进行评估，并对2次结果进行差异性分析以评估模型辅助效果。
结果本模型在内部测试集中对睫状体前位识别的准确率为93.43%，灵敏度为84.30%，特异度为97.78%。本模型在外部测试集中表现良好，准确率为92.81%。人机比赛结果显示，本模型识别的准确率与高年资眼科医师相近，并优于其中2名高年资眼科医师。3名高年资眼科医师平均总用时为726.73 s，模型分类总用时为58.30 s，高年资医师平均用时较长，约为模型分类的12.47倍。经模型辅助后，8名低年资医师诊断准确率为86.71%，明显高于辅助前的76.25%；评估总用时为(714.91±213.82)s，明显低于辅助前的(987.90±238.56)s，差异均有统计学意义( χ 2＝－7.550， P<0.001； t＝2.774， P<0.05)。
结论基于深度学习算法的UBM图像分析系统在PACG睫状体前位诊断中展现出较高的准确率，为低年资医师的UBM识图训练提供了有力支持。
ObjectiveTo explore the clinical application value of a deep learning algorithm-based ultrasound biomicroscopy (UBM) image analysis system for primary angles-closure glaucoma (PACG) anterior located ciliary body.
MethodsA diagnostic test study was conducted.A total of 2 132 UBM images from 726 eyes of 378 PACG patients who underwent UBM examination were collected at Renmin Hospital of Wuhan University from August 2022 to December 2023.The dataset was divided into a training set of 1 599 images and a test set of 533 images, and a deep learning algorithm was employed to construct a model.An additional 334 UBM images from 101 eyes of 69 PACG patients treated at Huangshi Aier Eye Hospital were selected to conduct external testing.A separate set of another 110 UBM images were selected for a human-machine competition to compare the accuracy and speed between anterior located ciliary body evaluation system and three senior ophthalmologists.Furthermore, eight junior ophthalmologists assessed the 110 UBM images independently without and with the assistance of the model, and the differences between the two evaluations were analyzed to assess the assisstance effect of the model.This study adhered to the Declaration of Helsinki.The study protocol was approved by the Ethics Committee of Renmin Hospital of Wuhan University (No.WDRY-2022-K109).
ResultsThe model achieved an accuracy of 93.43% for anterior located ciliary body identification in the internal test set, with a sensitivity of 84.30% and a specificity of 97.78%.The model also performed well on the external test set with an accuracy of 92.81%.In the human-machine competition, the model's accuracy was comparable to that of the senior ophthalmologists and outperformed two of the three senior ophthalmologists.The average total time of the three senior ophthalmologists was 726.73 seconds, approximately 12.47 times longer than the model's 58.30 seconds.With model assistance, the diagnostic accuracy of the eight junior ophthalmologists was 86.71%, which was significantly higher than 76.25% without model assistance ( χ 2=-7.550, P<0.001).And the image evaluation time was (714.91±213.82)seconds, which was significantly lower than (987.90±238.56)seconds without model assistance ( t=2.774, P<0.05).
ConclusionsThe UBM image analysis system based on a deep learning algorithm demonstrates high accuracy in diagnosing anterior located ciliary body in PACG and provides a strong support for the UBM image recognition training of junior ophthalmologists.
丛钰玉,蒋维艳,朱剑,等. 基于深度学习的原发性闭角型青光眼睫状体前位的自动评估[J]. 中华实验眼科杂志,2024,42(12)：1134-1141.
DOI:10.3760/cma.j.cn115989-20240328-00085版权归中华医学会所有。
未经授权，不得转载、摘编本刊文章，不得使用本刊的版式设计。
除非特别声明，本刊刊出的所有文章不代表中华医学会和本刊编委会的观点。
丛钰玉：设计试验、实施研究、采集数据、分析/解释数据、文章撰写；蒋维艳：设计试验、实施研究、采集数据、分析数据；朱剑：提供外部测试集数据；郑碧清：参与人工智能模型搭建及测试；杨燕宁：设计试验、文章审阅及定稿
文档即答
智问智答
机器翻译
你好，我可以帮助您更好的了解本文，请向我提问您关注的问题。
