CrossTransUnet: A New Computationally Inexpensive Tumor Segmentation Model for Brain MRI

Brain tumors are usually fatal diseases with low life expectancies due to the organs they affect, even if benign. Diagnosis and treatment of these challenging tasks, for experienced physicians experts, heterogeneity tumor cells. In recent years, advances in deep learning (DL) methods have been integrated aid diagnosis, detection, segmentation brain neoplasms. However, is a computationally expensive process, typically based on convolutional neural networks (CNNs) UNet framework. While has shown promising results, new models developments can be incorporated into conventional architecture improve performance. this research, we propose three new, inexpensive, inspired by Transformers. These designed 4-stage encoder-decoder structure implement our cross-attention model, along separable convolution layers, avoid loss dimensionality activation maps reduce computational cost while maintaining high The attention model different configurations modifying transition encoder, decoder blocks. proposed evaluated against classical network, showing that differences up an order magnitude number training parameters. Additionally, one outperforms UNet, achieving significantly less time Dice Similarity Coefficient (DSC) 94%, ensuring effectiveness segmentation.