Unsupervised Brain Tumor Detection
A fully automatic, unsupervised algorithm that can detect single and multiple tumors ranging in size from 3 to 28,079 mm, and has the potential to discriminate between suspicious and normal brains is presented.
Abstract
Accurate and fast automatic detection of brain tumors in 3D MR neuroimages can significantly aid early diagnosis, surgical planning, and follow-up assessment. Primary and metastatic tumors, which present substantial challenges to both human and state-of-the-art brain tumor detection algorithms due to their diverse location and size, need to be carefully monitored. We present a fully automatic, unsupervised algorithm that can detect single and multiple tumors ranging in size from 3 to 28,079 mm. Using 20 clinical 3D MR scans containing 1 to as many as 15 tumors per scan, our proposed approach achieves an 87.84 95.30% detection rate and an average end-to-end running time of 4 minutes. In addition, 5 normal clinical 3D MR scans are evaluated quantitatively to demonstrate that our approach also has the potential to discriminate between suspicious and normal brains.