基于U-Net的T细胞斑点检测方法研究

针对传统图像分割方法抗噪性弱、容易漏检的问题，提出基于U-Net模型的T细胞斑点分割算法。通过中值滤波器平滑消除噪声，灰度化处理降低背景干扰，采用Adam算法优化损失函数，能有效提高分割准确率。实验结果表明，与基于区域生长的传统分割方法对比，U-Net方法在少量斑点和较多斑点两种情况下F1分别提升9%和6%，验证了其有效性。

T-cell spot test based on U-Net

Abstract: Aiming at the problems of weak noise resistance and high probability of missed detection in traditional image segmentation method, a T-cell spot segmentation algorithm based on U-Net model is proposed. The segmentation accuracy is effectively improved by smoothing noises through median filter, reducing background interference by graying, and taking Adam algorithm as a loss function. The experimental results show that, compared with the traditional segmentation method based on region growth, the proposed method based on U-Net increases F1 by 9% and 6% in the experiments with a small number of spots and lots of spots, which verifies its effectiveness.