Improved MRI brain tumor instance segmentation model based on attention mechanism

To address the segmentation issues caused by the objective complexity of MRI brain tumor images, this paper proposes a novel instance segmentation model called Bi-CFpn network with an optimized neck. It adds a bottom-up enhancement path on the basis of FPN, and further introduces the attention mechanism into this enhanced path to form a more optimal attention coefficient for multi-scale tumor feature fusion. Experimental results show that the proposed can preserve the feature information of tumors in high-dimensional nonlinear spaces, enhances sensitivity to the spatial location of targets, and improves the accuracy of target detection and segmentation, while also takes computational efficiency into account.