Brain Tumor MRI Clustering Algorithm Based on Nearest Neighbor Prior Variational Autoencoder

Accurate detection of brain tumors is crucial for early diagnosis and treatment planning. However, the high-dimensional complexity of medical images, intra-tumor heterogeneity, and imaging noise limit the effectiveness of traditional clustering methods in MRI analysis. To address these challenges, this paper proposes an improved Variational Autoencoder (VAE)-based clustering algorithm that integrates knowledge self-distillation and a nearest neighbor prior to enhance feature learning and clustering semantic consistency. Additionally, we design medical image-specific data augmentation strategies to improve model robustness. Experimental results show that our method outperforms existing deep clustering algorithm based VAE in terms of sensitivity, specificity, and clustering accuracy for brain tumor MRI images. This work holds promise for applications in brain tumor detection and related medical tasks.