Brain Tumor Segmentation from MRI Scans Paper
This paper presents an exploration of tumor segmentation from MRI scans, focusing on the efficacy of thresholding and region growing methods, using manual and automatic seed selection. We employed two diverse datasets from the Brain Tumor Segmentation (BraTS) challenge to evaluate our approaches. We meticulously compare the performance of our methods against eight established metrics, encompassing both quantitative and qualitative dimensions. Quantitative analysis involves metrics such as Jaccard similarity and dice coefficient, accuracy, specificity, and runtime, providing a multi-faceted view of the segmentation performance. Qualitative analysis is conducted through visual inspection, ensuring that the segmented tumors align closely with the groundtruth labels. The results demonstrate a superiority of the region growing methods, especially with automatic seed selection, in accurately delineating tumor boundaries. This study contributes significantly to the field of medical image processing, offering insights that could be useful for tumor segmentation.
This project has been conducted as a part of my PhD at the University of Ottawa, Canada.