A method for predicting brain disease state change is disclosed. The method includes acquiring test images obtained by capturing a portion of a human brain at a time interval, performing a pre-processing procedure of converting the test images into test voxels configured to be processed for image analysis, wherein a respective test voxel of the test voxels is composed of three-dimensional voxel units, mapping first and second test voxels selected from the test voxels acquired from a patient, with each other on a three-dimensional voxel unit, wherein the first test voxel is acquired at a first time-point and the second test voxel is acquired at a second time-point, generating a voxel-based data-set based on the mapped first and second test voxels, extracting a change in the test voxels using a deep neural network, and generating a state change probability model based on the change in the test voxels.