U-Transformer Driven Brain Tumor Recurrence Prediction Using Radiomic Feature Extraction

The personalized treatment of brain tumor is accomplished with the prediction of brain tumor recurrence and has become a crucial element in neuro-oncology to enhance the patient’s health. Conventional methods utilize techniques including Convolutional Neural Networks, U-Net often failed to capture the intricate, long-range dependencies from the MRI images. The proposed work is based on the U transformer which predicts the recurrence of Brain tumor with the extraction of features of MRI images taken before and after the surgery of the patient. For robust analysis, radiomic features are extracted with the PyRadiomics. This is the combination of both Wavelet filtering and Laplacian of Gaussian Filtering. For selecting the features, the Modified Relief algorithm is employed and the features that are chosen from both the pre and post-surgery are fused with the weighted fusion framework. The images are pre-processed using the normalization technique and the U-Transformer will predict the likelihood of BT recurrence. The U-Transformer is the integration of transformer layers into the traditional U-Net framework. The prediction accuracy of the proposed brain tumor recurrence prediction is enhanced. The outcomes stated that the proposed U-transformer could be the power tool to predict the recurrence detection earlier and thereby significantly assist the physicians regarding the followup treatments for future enhancement.