Comparison of Tumor Segmentation Techniques from Medical Images

A Brain Tumor (BT) is one of the leading causes of mortality worldwide. Because of this, early diagnosis is crucial. BT localization and segmentation from magnetic resonance imaging (MRI) is a challenging but crucial task with many medical analytic applications. Many methods have been presented for tumor segmentation. However, they each have their issues, such as reliance on manual intervention from experts and uncertainty over which feature extractor to use. To overcome these obstacles, a novel, hybrid strategy for tumor region segmentation based on a convolutional neural network is suggested. Kaggle’s Brain MRI data was used for training and testing the proposed approach. First, a pre-processing strategy is used to operate on a selected portion of the image instead of the entire image to obtain a robust and adaptable BT segmentation system. Reduced processing time and no overfitting issues are the results of this strategy. Second, the Res-SegNet hybrid model was employed, now the model works only with a smaller portion of the MRI Accuracy, loss, Intersection of Union (IoU), and Dice coefficient are some of the metrics that were used to gauge how well the proposed method performed. When applied to tumor segmentation, the model achieved an accuracy of 0.99, an IoU of 0.745, a Dice of 0.831, and a loss of 0.176.