An Intelligent Diagnosis and Tumor Segmentation Method based on MRI Images Using Pre-trained Deep Convolutional Neural Networks (CNNs)

Abstract

The correct segmentation detection and classification of brain tumors in MRI images plays an essential role in discovering neurological problems at early stages and their subsequent treatment management. Deep learning has made Convolutional Neural Networks (CNNs) highly effective for complex medical imaging processing because they perform self-learning to find subtle features present within the data. Testing four CNN models for brain tumor segmentation was conducted using 1,251 MRI images from BraTS2021 dataset through CaPTk, 2DVNet, EnsembleNets and ResNet50 evaluation. The research relied on the combination of DSC and HD to perform quantitative analysis. The EnsembleUNets achieved the best performance since they produced the lowest HD of 18 and the highest DSC of 0.92. The suggested EnsembleUNets delivered exceptional capability through a CCC measurement of 0.75 and achieved the lowest RMSE value of 0.52 together with the highest TDI value of 1.9 for tumor segmentation and classification tasks in clinical scenarios. The obtained results indicate EnsembleUNets effectively segment brain tumors and identify their type as well as classify them in medical settings.