Regarding the 3D reconstruction method using medical images, the magnetic resonance (MR) imaging data were selected to reconstruct the 3D geometry of the brain parenchyma and the cerebral ventricles system in order to build up a preprocessing platform for the virtual physiological human (VPH) brain, e.g. the cerebroporomechanics model. Moreover, the progress of the 3D geometry scheme should divide into two steps, segmentation and reconstruction. Firstly, we obtain a speed image via a semi-automatic algorithm, which is based on Gaussian mixture model-based clustering and Supervised classification using random forests and utilise the difference between the foreground image and background image that the positive is defined by the region of interesting in the foreground image and the negative is defined by the region of interesting in the background image. Furthermore, it is necessary to label different tissues as markers. Each marker of the voxel will generate the eigenvector, which could be a pixel intensity or space coordinates. Afterwards, by using the classification method does the clustering based on a machine learning approach via the feature that we choose. Additionally, the scheme of the 3D medical images reconstruction includes to find the contour on the target area and to use the active contour model and the initialisation seed. The active contour model will be an adaptive classification of the region of interesting according to the speed image and the initialisation seed that we choose. Within this manner, the 3D geometry should be obtained via the reconstruction scheme. The following pictures demonstrate the 3D human brain and rat brain via the reconstruction scheme.