Alzheimer’s Disease (AD) is a neurodegenerative disorder which progresses gradually and worsens over a period resulting in cognitive decline and affecting the motor skills of the person. As there is no cure for dementia, it is imperative to seek the assistance of medical professionals to curtail its progression. Further, accurate diagnosis at an early stage is very much essential to increasing the lifespan of the patients. Various Deep Learning techniques have recently been employed as promising tools for faster and more accurate detection and classification. This paper proposes a novel custom 3D CNN architecture based on a 3D extension of LeNet. This method ensures a more lightweight and quicker binary classification of Alzheimer’s disease even on a small dataset. 3D MRI images from the ADNI dataset were used for training and testing. The proposed custom 3D CNN is compared with the existing 3D LeNet and 3D UNet models in terms of accuracy, loss, time per epoch, sensitivity, and ROC AUC. It is found to outperform the existing architectures.