The brain is an essential part of the entire nervous system because it can control and regulate body parts. The International Agency for Research on Cancer (IARC) estimates that 76% of deaths from brain tumors are cancer related. Abnormal cells in the brain cause brain tumors and make life dangerous if untreated at the early stages. A significant portion of the Image is extracted using image processing techniques during tumor image classification. The 3D shape is a complex structure in the 3D space; because of this reason, there would be a minimal of 3D shapes for feature learning. To cope with those challenges, research paper transform 3D shapes into 2D space and use automatic encoders to learn features from 2D images. The proposed 3D based Spatial Auto Encoder (3D-SpAE) approach is automated by learning a state representation directly from camera images and the state-space constructions. Once the autoencoder is trained, the coefficients acquired for reconstructing an image based on prototypes which are utilized as a feature for 3D shape matching and retrieval. The autoencoder can achieve high performance for image retrieval because it can learn feature adaptively from training data. 3D-SpAE classified the brain tumor data into Whole-Tumor, Tumor Core, and Enhancing Tumor. The results showed that the suggested method obtained better values of the accuracy of 99.4% when compared to the existing Nonparametric Localization Enhancement Methods with U-Net, 3 D Context Deep Supervised U-Net, Deep Elman Neural network with adaptive fuzzy clustering, NSGA, Convolutional Neural Network, and Kernel-based SVM.