You can start with FCM clustering for segmentation and evaluate the segmentation accuracy for the ROI. Then you can try other FCM algorithms based on trial and error basis. Basic FCM clustering is sensitive to noise and hence for noise corrupted images spatial-FCM can be experimented

Better start with traditional FCM method. Then try to implement the FCM within the ROI.

Modifications can be done for the following:

1) Automatically select the number of regions to cluster

2) Automatically initialize the membership functions for the selected clusters

3) Modify the objective function

These are only possible with the application knowledge, here MRI characteristics of brain and brain anatomy.

This supports to converge quickly

You can start with Conventional Fuzzy C-Means. Since traditional FCM is sensitive to noise and it ignores the neighborhood information you can use Kernel Spatial FCM method.

In addition you can also explore advanced fuzzy clustering methods like Intuitionistic Fuzzy C-Means methods. You can refer following papers

https://www.researchgate.net/publication/316589301_A_Modified_Intuitionistic_Fuzzy_Clustering_Algorithm_for_Medical_Image_Segmentation

https://www.researchgate.net/publication/320195713_A_novel_fuzzy_clustering_based_system_for_medical_image_segmentation