A Tumor MRI Image Segmentation Framework Based on Class-Correlation Pattern Aggregation in Medical Decision-Making System

Medical image analysis methods have been applied to clinical scenarios of tumor diagnosis and treatment. Many studies attempted optimize the effectiveness MRI segmentation by deep learning, but they do not consider optimization local details interaction global semantic information. Second, although medical pattern recognition can learn representative features, it is challenging ignore useless features in order generalizable embeddings. Thus, a tumor-assisted method proposed detect lesion regions boundaries with complex shapes. Specifically, we introduce denoising convolutional autoencoder (DCAE) for noise reduction. Furthermore, design novel framework (NFSR-U-Net) based on class-correlation aggregation, which first aggregates patterns images form class-correlational representation. Then relationship similar class identified closely correlate dense representations classification, conducive identifying data high heterogeneity. Meanwhile, model uses spatial attention mechanism residual structure extract effective information dimension enhance statistical images, bridges gap skip connections. In study, over 4000 from Monash University Research Center Artificial Intelligence are analyzed. The results show that achieves accuracy up 96% low resource consumption.