In recent years, numerous studies have established individual or combined imaging modalities as fundamental approaches for the early diagnosis of Alzheimer’s disease (AD). Additionally, cortical surface analysis has proven effective in brain disorder research due to its ability to depict brain structure. This study introduces a novel multimodal methodology that integrates information from both imaging and surface data for AD diagnosis. Specifically, we leverage knowledge from surface representation to enhance the performance of multimodal image-based classification, streamlining the inference process by relying solely on image modalities. Furthermore, we have designed a multi-view, multi-convolution block to improve the representation capabilities of image extractors. We demonstrate the reliability of the proposed method using the entire ADNI series dataset, including ADNI1, ADNI2, and ADNI3, with comparison to previous studies that focused solely on ADNI1. Our method shows promising results in early diagnosis of AD, surpassing conventional approaches with an \(80\%\) balanced accuracy (BACC) for ADNI1. We validate the robustness of the proposed method on the ADNI2 and ADNI3 datasets, which incorporate more contemporary PET modalities for brain disease, demonstrating its consistent performance with a result of \(79\%\) BACC.