Aiming at the problem that the subtle features in magnetic resonance images ( MRI) are difficult to ex- tract and easy to be missing , a method for extracting subtle features of magnetic resonance images based on the fusion of Canny-SIFT algorithm is proposed. The algorithm first solves the problem of unclear texture and subtle feature infor- mation in the image due to the uneven gray level of the image and the complex noise signal. It uses automatic Gamma transformation to increase the image contrast. Separately deal with the noise of the observation area; for the incomplete extraction of subtle features in the key observation area , the importance is divided according to the actual diagnosis re- quirements , and the location , area and other information of the important area are obtained through feature matching and topological relationship reasoning , and the self-adaptation is completed at the same time Select the corresponding Sobel operator; finally , the output image is obtained by thresholding segmentation and binarization ; Experiments show that the proposed method has significantly improved the edge detection accuracy compared with the existing Canny method , the structural similarity is increased by 38% , and the mean squared error is reduced by 31. 4% , and the per- formance is the best compared with other mentioned algorithms.