Mai WangXiang, He WeiYi, Mo Rongchi, Liu GuoHao, Hong Jing, Li WanYue, Luo Li, Chen ZhuoMing
The First Affiliated Hospital of Jinan University, Department of Rehabilitation, Guangzhou, China.
The First Affiliated Hospital of Jinan University, Department of Pediatrics, Guangzhou, China.
J Biomed Opt. 2025 Jun;30(6):065003. doi: 10.1117/1.JBO.30.6.065003. Epub 2025 Jun 30.
Platelet-rich plasma (PRP) is a critical component in regenerative medicine, with applications in tissue repair and inflammation regulation. Consistent preparation quality is essential for therapeutic efficacy, but traditional quality control (QC) methods are labor-intensive, slow, and prone to variability.
We introduce a computer vision-based automated PRP QC model using deep learning to improve the efficiency and accuracy of PRP preparation.
Blood samples were collected and processed in the laboratory to prepare PRP. Images of the samples were manually captured. Medical-grade QC evaluations determined sample quality, which was labeled for model training. The image data were preprocessed and analyzed using a ResNet18 convolutional neural network combined with a binary classifier to develop a PRP QC model. Training and testing were conducted using data from patients, and the model's accuracy was tested on the independent unavailable dataset.
The PRP QC model achieved an average classification accuracy of 82.5% on unavailable datasets (previously unseen test samples), significantly reducing the time required for QC to under 1 min.
We demonstrate a nondestructive, real-time QC method for PRP preparation with computer vision and deep learning, offering a practical and scalable solution to improve clinical outcomes in regenerative medicine.
富血小板血浆（PRP）是再生医学中的关键组成部分，可用于组织修复和炎症调节。一致的制备质量对于治疗效果至关重要，但传统的质量控制（QC）方法劳动强度大、速度慢且容易出现变异性。
我们引入一种基于计算机视觉的自动化PRP质量控制模型，利用深度学习来提高PRP制备的效率和准确性。
在实验室采集血样并进行处理以制备PRP。手动采集样本图像。医学级质量控制评估确定样本质量，并对其进行标记以用于模型训练。使用ResNet18卷积神经网络结合二元分类器对图像数据进行预处理和分析，以开发PRP质量控制模型。使用患者数据进行训练和测试，并在独立的不可用数据集上测试模型的准确性。
PRP质量控制模型在不可用数据集（以前未见过的测试样本）上实现了82.5%的平均分类准确率，显著将质量控制所需时间缩短至1分钟以内。
我们展示了一种利用计算机视觉和深度学习对PRP制备进行无损、实时质量控制的方法，为改善再生医学的临床结果提供了一种实用且可扩展的解决方案。
