In this paper, a content-based scheme for assisting collecting similar medical image to be used for the tutorial of differential diagnosis is presented. The system uses visual-based user interface to allow the user to enter or query an image by selecting the region of interest (ROI) regions; and uses neural network to classify the relationship between the images stored in database. The system will output a set of candidate images that are texturalsimilar to the query image. If there are more than one input blocks, the system will perform multiple query and then combining the result set to obtain the most similar one to output. We extract the major 2D FFT coefficients to represent the texture features of each marked region and uses Kohonen self-organizing network to classify those extracted FFT coefficients. In the query stage, the system first individually checks which texture category is the query block in then uses intersection and some criteria to identify the most likely candidate image. The experimental results show that in simple query, 96.6% of images can be correctly retrieved with the displacement up to 22% of the block size. Based on this result, when user input more than one block, the impact of the shift variation due to human marking would be small.