目的探讨CT图像智能识别的 肺结核(TB)小点的可行性。方法 收集 100张CT图像，通过图像预处理、图像分 割、图像分类技术以识别TB小点。结果 100张CT图像中，54张有肺结核腔，46张 没有肺结核腔。数据显示混合分类法最 佳，FPR为0.144/图，且速率可达当使用 逆梯度的变异系数(GICOV)单独或循环 方法时的两倍。基于不同的分类两个CT 测试图像的比较发现，混合方法性能更 优。结论 利用CT图像分割技术进行TB小 点的诊断，可获得可靠、准确的诊断结 果。

Objective To investigate the feasibility of CT image intelligent recognition of tuberculosis (TB) dot. Methods 100 CT images were selected to identify TB point through image processing, image segmentation and image classification technology. Results Among the 100 images, there were 54 pieces tuberculosis cavity and 46 no pulmonary tuberculosis cavity. Data showed that mix was the best way, FPR was 0.144 /figure, and its speed could reach to two times of that using inverse gradient variation coefficient (GICOV) alone or circulation method. Based on the classification of different two CT test image comparison, the mix method had a good performance. Conclusion CT technology to identify TB dot, in order to obtain precise and accurate diagnosis