Page 30 - IJEEE-2022-Vol18-ISSUE-1

26 | Atiyah & Ali

Approach Model TABLE I Precision Recall Specificity
Region Based U-Net-Res The Performance Metrics Summary 0. 9807 0. 9886 0. 9935
Segmentation Dice Loss IoU Score F1 Score Accuracy
Edge Based Net50 0. 009538 0. 7375 0. 9846 0. 9923 0. 9852 0. 9888 0. 9951
Segmentation U-Net-
ResNet50 0. 008768 0. 7542 0. 9870 0. 9935

Fig. 6: The learning curve of Edge-based segmentation Fig. 7: The learning curve of region-based segmentation

Fig. 8: Prediction after edge-based segmentation Fig. 9: Prediction after region-based segmentation

TABLE II

Comparison of The Proposed Method with Previous Works

Author Dice Score Accuracy

Ramin et al.[17] 0.9203 -

Gunasekara et al.[10] 0.92 0.9457

Fabian et al.[9] 0.8895 -

Chinmayi et al.[5] - 0.9801

Xue Feng et al.[8] 0.926 -

Proposed Method 0. 9870 0. 9935

V. CONCLUSIONS model is evaluated based on dice loss, IoU score, f1 score,
precision, recall, accuracy, and specificity. The edge-based
Various approaches for brain tumors segment were segmentation achieved the highest scores in all performance
presented in this paper. Firstly, using the technique of edge- metrics with the dice loss score of 0. 008768, IoU score of 0.
based segmentation, and other using region-based 7542, f1 score of 0. 9870, the accuracy of 0. 9935, the
segmentation. Here, proposed a U-Net with ResNet50 precision of 0. 9852, recall of 0. 9888, and specificity of 0.
encoder architecture for efficiently segmenting brain tumors. 9951. The edge-based segmentation played a key role in the
In this study comparative analysis of region-based treatment of brain tumors, according to the test findings. The
segmentation and edge-based segmentation using U-Net proposed model foresees the segmentation of brain injuries
with ResNet50 encoder, architecture is performed. The and assists in the precise segmentation of the lesions.