The Added Value of Automatic Segmentation of Cerebral Gliomas in Multi-Sequence Magnetic Resonance Imaging (MRI)

Gliomas are the most common primary brain tumors and are classified by their histopathological appearances using the World Health Organization (WHO) system into low-grade glioma (LGG) (grades I and II) and high-grade glioma (grade III anaplastic glioma and grade IV glioblastoma.

Gliomas, particularly high-grade, exhibit irregular growth patterns infiltrating the surrounding brain and thus showing irregular boundaries that may not be clear on conventional magnetic resonance images (MRI) MR images are visually inspected by radiologists, however, visual assessment is subjective, time consuming and prone to variability due to inter-rater differences. Accurate delineation of tumor boundaries as well as assessment of tumor volume are essential for treatment planning and monitoring treatment response . However, accurate delineation of the boundaries of glioma using subjective visual assessment is often difficult due to tumor heterogeneity and complexity, overlapping signal intensity with surrounding tissues and uneven tumor growth into nearby structures .

Compared to tumor volumetry, the routine visual evaluation of tumor size is based upon simple linear measurements of the gross tumor volume. These bi-dimensional measurements are often performed on a single MRI slice without volumetric measurements. These linear measurements are user-dependent and prone to errors due to increased measurement variability, especially in irregularly shaped lesions Computer-based fully-automatic tumor segmentation methods present a possible solution to these issues. The process is based upon information extraction from structural brain MRI images using a probabilistic tissue model to define the clear tumor boundaries using different MRI pulse sequences. These methods could accurately and rapidly identify glioma from surrounding normal brain tissue, and perform tumor volumetry, while eliminating intra-observer and inter-observer variability Internal changes within glioma, such as enhancement pattern and degeneration are crucial for identification of glioma grade, planning of treatment, monitoring of disease progression and evaluating the efficacy of therapy. In the process of automatic glioma segmentation, different parts of the glioma are characterized as solid (active) tumor, necrosis and peri-tumoral edema .

Automatic segmentation methods utilize artificial intelligence and machine learning techniques for extraction of information from multi-sequence MRI including, basically, T1W, Gadolinium enhanced T1W, T2W and FLAIR sequences .

Appropriate assessment of the extent of tumor resection plays an important role in the prognosis of glioma, since maximizing the extent of resection influences survival in these patients. Complete resection of enhancing tumor, defined as the removal of the final 1-2% of the tumor, seems to provide the most benefit in terms of patient's survival . Automatic segmentation could lead to better diagnosis and proper treatment planning through accurate tumor localization and classification .