Discrepancies between frame- and CBCT-based stereotactic space definition on the Gamma Knife Icon.

PURPOSE: To assess differences between frame-based and cone beam computed tomography (CBCT)-defined stereotactic space and to identify predictors of the observed findings. METHODS AND MATERIALS: Differences between frame-based and CBCT-defined stereotactic space after image co-registration were reviewed for 529 patients. Treatment planning system reported the information about the shifts in X, Y, and Z coordinates of the center of the stereotactic space (i.e., coordinate X = 100 mm, Y = 100 mm, and Z = 100 mm) defined by the frame, and the maximum shot displacement (MSD) in mm. We collected the potential predictors of the differences. In total, 19 factors were investigated. We used multiple linear regression to evaluate associations with the increased differences. RESULTS: Rotational and translational shifts greater than 1° and 1 mm, respectively, were observed in 2.6% of patients. At the same time, a decrease in tumor coverage of more than 5% was detected in 8.3% of cases. It was revealed that the higher fiducial errors (both mean and maximum), the greater weight of the patient, and the lower Karnofsky Performance Scale were predictors of increased rotational, translational shifts, and the MSD.

The last decade's experience of management of central neurocytomas: Treatment strategies and new options.

BACKGROUND: The purpose of the presented work is to evaluate the last decade's experience in surgical management of central neurocytoma (CN) and elucidate on the treatment strategies and new options. METHODS: The current series consists of the remaining 125 patients (70 females and 55 males) operated on during the past decade from 2008 to 2018. Most tumors were resected through transcortical (n = 76, 61%), or transcallosal (n = 40, 32%) approaches. In 5 (4%) patients with predominantly posterior location of the tumor, non-dominant superior parietal lobule approach was utilized. Both approaches (transcortical + transcallosal) were used in 4 (3%) of cases. Seven consecutive patients with large CN underwent prophylactic intraventricular stenting to prevent hydrocephalus. RESULTS: Gross total resection was achieved in 45 patients (36%), subtotal resection (STR) in 40 (32%) cases. After surgery, 63 (50%) patients had neurocognitive problems, including disorientation, attention deficit, global amnesia, short-term memory deficits, and perceptual motor and social cognition problems. A total of 26 patients (21%) had postoperative hemorrhage in the resection bed. Obstructive hydrocephalus was noted in 25 (20%) patients. The entrapment of the occipital and/or temporal horns was observed in seven cases. None of the seven patients with prophylactic intraventricular stents required shunting. CONCLUSION: Although high rates of gross total or STR can be expected, the mortality and morbidity remain significant even in the modern neurosurgical era. Prophylactic intraventricular stenting in patients with large posteriorly located tumors with hydrocephalus may prevent ventricular entrapment and shunting. The main risk factors for recurrence are presence of residual disease and Ki-67 index over 5%. Recurrent symptomatic tumors should be treated surgically, whereas asymptomatic progression can be managed with stereotactic radiosurgery. Both treatment modalities are associated with low risk of complications and high tumor control rates.

Challenges in Building of Deep Learning Models for Glioblastoma Segmentation: Evidence from Clinical Data.

In this article, we compare the performance of a state-of-the-art segmentation network (UNet) on two different glioblastoma (GB) segmentation datasets. Our experiments show that the same training procedure yields almost twice as bad results on the retrospective clinical data compared to the BraTS challenge data (in terms of Dice score). We discuss possible reasons for such an outcome, including inter-rater variability and high variability in magnetic resonance imaging (MRI) scanners and scanner settings. The high performance of segmentation models, demonstrated on preselected imaging data, does not bring the community closer to using these algorithms in clinical settings. We believe that a clinically applicable deep learning architecture requires a shift from unified datasets to heterogeneous data.