This work is motivated by breast cancer imaging data, which is quite unique in that the signals of tumor-associated microvesicles (TMVs) are randomly distributed from each individual image. This imposes a significant challenge for conventional imaging regression and dimension reduction models assuming a homogeneous feature structure. We develop an innovative multilayer tensor learning method to predict disease status effectively through utilizing subject-wise imaging features and integrating multimodality information. Specifically, we construct a multilayer decomposition which leverages an individualized imaging layer in addition to a modality-specific tensor decomposition. One major advantage of our approach is that we are able to efficiently capture the individual-specific spatial information of microvesicles through combining multimodality imaging data as well as incorporating modality-wise features simultaneously. To achieve scalable computing, we develop a new bi-level block improvement algorithm. In theory, we investigate both the algorithm convergence property and asymptotic consistency for model estimation. We also apply the proposed method for simulated and human breast cancer imaging data. Numerical results demonstrate that the proposed method outperforms other existing competing methods. This is joint work with Xiwei Tang and Xuan Bi.