Positron emission tomography (PET) scans have emerged as a widely accepted tool for diagnosis of Alzheimer’s disease (AD) using machine learning and its development stage. Existing models utilize MRI data to evaluate and forecast disease progression and medical conditions of the brain ultimately detecting the existence of Alzheimer’s disease. In this study, we predict brain disease prognosis using a weakly supervised deep neural network (wiseDNN) and a landmark detection algorithm to facilitate the prediction and prognosis of AD using PET images. The model gives better performance compared to existing models and incorporates the incomplete ground-truth values which is clinical measures from four different visit-points with missing values to facilitate prediction of the disease, and by extension, its prognosis at future visit-points. Our proposed deep neural network for disease prognosis leverages joint PET data and missing ground-truth clinical scores of subjects at multiple visit-points. By integrating PET and clinical measures, we aim to enhance the precision of prediction models. Overall, our study demonstrates the potential of utilizing PET data and deep neural networks to accurately diagnose AD and its progression. The use of landmark detection and joint PET and clinical measures enhances the precision of prediction models, positioning our approach as a promising tool to aid in diagnosing and managing AD.