Evaluation of postoperative resection cavity MRIfindings of grade-4 glial tumors with semi-automatic tissue analysis method

Despite the clinical use of advanced imaging modalities in the post-surgical resection cavity follow-up of grade-4 glial tumors, there is still an ongoing challenges to differentiate the postoperative cavity related findings ['progression', 'pseudoprogression/radiation necrosis (CRT)']. Although the difference in experience and interpretation is among the reasons for this difficulty, advanced imaging methods cannot be used routinely in follow-up imaging due to their high cost. The aim of this study was to demonstrate the potential of differentiating surgical cavity and treatment-related imaging findings using texture features derived from radiological imaging techniques. This differentiation can be achieved without the need for advanced imaging methods like perfusion MRI or MR spectroscopy. In addition, this distinction can be made using texture features of conventional sequences and diffusion and magnetic susceptibility imaging, which have been routinely used, without the need for advanced imaging methods such as perfusion MRI, MR-spectroscopy. Adult patients (age>18) who were diagnosed with grade-4 glial tumors (Glioblastoma, IDH-wild type and astrocytoma, IDH-mutant) and whose follow-up MRIs were performed in our clinic between July 2018 and 2022 after chemoradiotherapy were included in the study. Pathological areas in brain MRIs were semi-automatically segmentation using the "OLEA SPHERE" program, based on two different sequences (FLAIR and post-contrast T1-weighted sequences). These segmentations were overlaid onto five different sequences, and features were extracted from a total of 10 sequences. Information Gain method was used to identify features with high significance. In order to distinguish the treatment related manifestations in the surgical cavity that reflected in imaging as 'stable/tumor-free cavity', 'pseudoprogression/radiation necrosis (CRT)', and 'progression', a main data model was created by combining high-significance features obtained from each sequence with clinical variables (age-gender) and tumor molecular (IDH status) biomarkers of the patients. For classification of surgical cavity findings, the pathology results were utilized for patients with available pathology, while follow-up imaging evaluations were performed for patients without pathology. The main data model was built using data from 108 patients, and validation was conducted using data from 68 patients. Hybrid methods were employed to improve the classification performance of models created for each sequence (ADC, FLAIR, cT1-w, MDG [SWAN/SWI], MAGIC T1-w). The study demonstrated that treatment-related surgical cavity imaging findings could be differentiated with a highest accuracy classification rate (ACR) of 75%. Models created based on the rCT1A sequence, serving as a reference, exhibited better performance, possibly due to the selected features better reflecting the tissue characteristics of contrast-enhancing areas. Keywords: glioblastoma, grade-4 astrocytoma, texture analysis, IDH-mutant, IDH-wild