<p>In this study, we propose an ensemble clustering multiblock sparse multivariable analysis method for identifying disease subtypes using multiple (multimodal) brain images with multiple measurements or processing methods and voxel values as input data. First, a multiblock scoring method comprising upper and lower layers was implemented to reduce the image data. In the lower layer, principal component scores were computed for each modality. Subsequently, a new matrix was generated in the upper layer by summarizing the scores calculated in the lower layer, and clustering was conducted using non-negative matrix factorization. The lower layer scoring produced multiple components, while the upper layer yielded multiple clustering outcomes, which were subsequently enumerated. The usefulness of the proposed method is demonstrated using real brain disease data. Multimodal brain images offer a more robust data-driven subclassification of diseases, which is beneficial for clinical applications. Furthermore, the proposed method can be utilized to forecast the efficacy and side effects of treatment and is valuable for high-dimensional data analysis beyond brain imaging. This method can offer more comprehensive information and significant insights into disease subclassification.</p>