Comparison of MRI Image Segmentation Methods for Radiation-Induced Brain Edema After Radiotherapy for Patients with Intracranial Tumors

When patients receive stereotactic radiosurgery (SRS) treatment, more intracranial tumor cells usually need to be eliminated, and high radiation doses are required to improve tumor control rates. However, high doses can cause damage to the normal tissues around the intracranial tumor, resulting in normal tissues being unable to repair, and this leads to the complication of radiation-induced brain edema. This study examines the feasibility and relative effectiveness of three different image segmentation algorithms, as applied to the evaluation of hydrocephalus initiated by the radiative treatment of patients with intracranial tumors. We implemented three different segmentation algorithms to automate the tagging of brain edema regions, namely watershed transform segmentation, K-mean segmentation, and region growing segmentation. Two performance indices, Dice and Jaccard, were employed in our experiments. A performance index close to 1 is an indication of good correlation. Experimental results reveal that the region growing segmentation outperforms the other two. The region growing segmentation, K-mean segmentation, and watershed transform segmentation have Dice index values of 0.77, 0.72, and 0.71, respectively, and have Jaccard index values of 0.60, 0.56, and 0.55, respectively. Among the subjects of this comparative study, the region growing segmentation is superior in resembling the marking of brain edema suggested by expert clinicians.