针对胎盘超声图像自动分级这一临床应用问题，提出一种基于自适应多神经网络的分级算法。该算法与一般的一次性分离算法不同，其是通过设计两级BP神经网络模型来对胎盘图像进行两级分离。该算法在神经网络的训练中，对神经网络的输出没有采用一般的四舍五人来得到胎盘级数，而是采用了更合理的胎盘级数判定准则，并由此提出了一种自适应确定阈值的算法，用来判定胎盘级数。实验及临床应用表明：该算法能得到与专家手工分级基本吻合的自动分级结果，其阈值分割前得出的分级结果更可以给医生一个精确的定量衡量胎盘成熟期的参考，因此具有较好的临床应用前景。

To solve the problem of automatic classification for ultrasound placenta images, we put forward an algorithm based on adaptive multiple neural networks. Two layers of BP net models were designed to carry Two-Stage separation of the placenta in this algorithm other than general one stage separation algorithm. When training networks, we do not adopt the common method which rounds the output of the networks, but propose a more reasonable grading rule, then present an adaptive threshold-gotten method to determine the reasonable placenta level. Experiments and clinic applications indicate that the similar classification result can be gotten by our algorithm as by experts, and the classification result by the algorithm before threshold division can give the doctor a good reference on the precise measurement of the placenta maturity, thereby, it has a good future in clinic applications.