Abstract: A classification method of ultrasonic liver image is studied based on threshold segmentation.Under the condition of minimizing the false-negative probability，the confidence intervals of gray levels which are normal distributions in normal and fatty liver images can be obtained by minimizing the false-positive rate.Then classify the images by comparing the segmented images with the interval thresholds.The experimental results show that the classification rate of normal livers reaches to 100% for testing or training data，the classification rate of fatty livers reaches to 96.3% for testing or training data.Compared with the Back Propagation Neural Network（BP NN） method，the method costs less time than BP NN method when they have the same performance.

Key words: fatty liver, ultrasonic liver image, classification, threshold segmentation

摘要： 研究了一种基于阈值分割的脂肪肝超声图像分类方法。该方法在最小化假阴性错误率的基础上尽量减小假阳性错误，获得服从正态分布的正常肝和脂肪肝图像灰度的置信区间。利用各自的区间阈值对超声图像进行分割，通过比较两幅分割图像进行分类。实验验证表明，不论是测试数据还是训练数据，该方法对正常肝正确识别率均为100%，对脂肪肝正确识别率均为96.3%。与BP神经网络方法比较表明，该方法在分类准确性上与神经网络方法相当，但比BP网络方法花费了更少的时间。

关键词: 脂肪肝, 超声肝图像, 分类, 阈值分割