Accurate deep learning-based detection of nasopharyngeal carcinoma (NPC) magnetic resonance (MR) images is conducive to diagnosis and treatment. These images are characterized by high dimensionality, complex noise interference, and blurred tissue structure boundaries. How to extract key pathological features from massive imaging information and provide quantitative basis for clinical diagnosis remains an important challenge in the current field of medical image processing. This paper uses multi-window fusion technology to map multiple key window information to the pseudo-color space, realizing the integration of multi-dimensional feature information and compensating for the information limitations of single-window imaging. Experiments show that this method can effectively improve model accuracy.