The invention relates to medical image processing, and provides an IMT (Intima Media Thickness) measurement algorithm with good overall performance. By using the method, there is no need for artificial participation, the time complexity is relatively low, the result of automatic measurement and the result of manual measurement by experts are highly consistent, and a satisfactory measurement result can be obtained for intima media of different morphological structures. An ultrasonic carotid artery intima media thickness measurement method based on deep learning of the invention comprises the following steps: (1) ROI (Region of Internet) acquisition; automatically identifying the far end of an ultrasonic carotid artery by use of a convolution neural network (CNN), and extracting an ROI; (2) pixel classification: performing more advanced feature extraction, and after feature extraction, building a pattern classifier by use of a stacked auto-encoder and a logistic regression classification layer to complete image pixel classification; and (3) boundary extraction: selecting a classification region based on the area information and location of a target classification region, and performing polynomial curve fitting on the classification boundary. The method is mainly applied in medical image processing.