75. Jahrestagung der Deutschen Gesellschaft für Neurochirurgie

Prof. Dr.

Vorsitz

PD Dr.

Prof.

Comparison of tractography algorithms in glioma patients

Tumor

Vergleich von Traktografie-Algorithmen bei Gliompatienten

Tractography enables us to reconstruct white matter (wm) fibre bundles from MRI, providing crucial insights into anatomical structures. It has become essential for planning and risk stratification in neurosurgery. Peritumoral tractography is challenging due to arising edema, demyelination and infiltration, yet vital in patients with gliomas in the left perisylvian area, possibly causing aphasia. Several deterministic and probabilistic algorithms have been developed, that model wm based on tensors (DTI) or spherical harmonic functions (CSD). Here we compared tractograms derived by common algorithms, examining how they differ in patients with perisylvian gliomas.We included n=62 presurgical patients with left perisylvian gliomas (24 females, 38 males, average age=53,5±15.4, age range 21–83). All were right-handed and had an initial diagnosis of supratentorial, unilateral WHO grade II (6), III (20), or IV (36) glioma. We derived the language relevant tractograms, AF, IFOF, ILF and UF, on both hemispheres employing DTI- (tensor_det, tensor_prob) and CSD-based (SD_stream, iFOD2) algorithms via TractSeg. We compared the results by streamline length, volume and dice scores.DTI- and CSD-algorithms failed in 16% and 13% peritumorally, respectively, and in 2% and 0% on the healthy hemisphere. Streamlines in the pathologic side were significantly 3% shorter and tract volumes 10% smaller than on the healthy one for all algorithms. Tract volumes vary a lot for different algorithms, but their ratios are similar in both healthy and pathologic hemisphere. Using the largest volumes generated by iFOD2 as reference, we find the ratios 0.47, 0.33 and 0.28 for SD_stream, tensor_prob and tensor_det, respectively. The dice scores for the pathological side revealed that DTI tractograms are more similar (0.85±0.10) than CSD-based ones (0.51±0.10). iFOD2 tracts differ substantially from tensor_det tracts (0.38±0.10).Tractograms generated by different algorithms vary in size and shape. These variations are similar in healthy and pathological hemispheres. CSD-based algorithms are less likely to fail and, thus, are useful in neurosurgical practice. Since DTI-based algorithms tend to underestimate and iFOD2 to overestimate tracts, integrating both in clinical practice may offer more precise results, finding a middle ground.

Abstractvortrag | Abstract talk

Autofluorescence based recognition of brain tumors with a convolutional neural network

Autofluoreszenzbasierte Erkennung von Hirntumoren mit einem neuronalen Netzwerk

The delineation of malignant brain tumors from surrounding tissue is crucial for maximal safe resection. Existing tools for intraoperative tumor visualization lack the possibility of real-time tissue classification. We therefore built a coherent fiber bundle (CFB) endoscope for autofluorescence imaging and trained a convolutional neural network (CNN) to distinguish tumor tissue (T) and non-tumor tissue (nT) in an ex vivo proof of concept study.The T-cohort included frozen tissue sections of 35 glioblastoma and 30 brain metastasis patients (10 breast cancer, 10 melanoma, 10 lung cancer). The nT-cohort comprised frozen tissue sections of hippocampi and temporal lobes from 23 patients undergoing surgery for epilepsy treatment. For image acquisition, a laser with excitation wavelength of 473 nm was coupled into a CFB and tissue autofluorescence between 500-550 nm was collected using a bandpass filter. The acquisition time of the camera was set to 700 ms. A CNN (VGG-19) was trained to differentiate the labels T from nT wherein the ground truth was defined by histopathological diagnosis. Training, validation and test sets were randomly generated across all patients in 66.6% / 16.6% / 16.6% proportion. Set generation was repeated 5 times and reported results for the independent test sets are mean values of all iterations.In total, 3259 CFB images were acquired (T:2408, nT:851), with usually 40 CFB images per sample. Removal of CFB background was not improving classification accuracy and thus, was not performed. The classification accuracy of the CNN with a threshold probability of 0.5 for T-label assignment was 0.986 (sensitivity: 0.988, specificity: 0.982) for analysis of images and 0.98 (sensitivity: 0.971, specificity: 1) when calculating the result for each patient. To simulate clinical application where correct identification of healthy tissue is crucial, we created a specificity-tuned CNN with adjusted threshold probability of 0.9 for T-label assignment. This led to a decreased accuracy of 0.945 but increased specificity on the image evaluation approach to 1 (sensitivity: 0.92).Autofluorescence imaging through a CFB is capable of ex vivo recognition of brain tumor tissue with a CNN. The simplicity of the setupallows for intraoperative implementation and potentially enables real-time in vivo brain tumor delineation. To realize intraoperative transition, analysis of ex vivo bulk tissue will be upcoming to capture the signal complexity caused by tissue depth.

Network analysis of perisylvian glioma patients

Netzwerkanalyse bei perisylvischen Gliompatienten 

Gliomas can impair the white matter (WM) network, potentially resulting in dysfunction. We analysed the gliomas" effects on structural connectivity in relation to aphasia. We used the patients" connectomes to identify network impairments related to aphasia and network topology using network theory. We studied 47 patients with left-perisylvian gliomas (WHO grade II (6), III (15), IV (26), 32; females; mean age=53, SD=14; age-range=28-78) for aphasia assessment using the Aachen aphasia test (AAT). AAT subtests included a token test, naming, repetition, and speech comprehension. 15 patients presented with aphasia. We constructed whole brain tractograms with constrained spherical deconvolution (CSD), performed anatomically constrained tractography and CSD-informed filtering of tractograms using MRtrix3. The resulting connectomes were based on the Desikan-Killiany-Tourville atlas and FreeSurfer. We calculated graph-based measures, such as assortativity, local & global efficiency, and node measures like node strength, betweenness centrality, and clustering coefficients, comparing these between hemispheres and correlating with AAT results. We found significant differences between the two hemispheres in assortativity (p=3.8e-5) and global efficiency (p=3.6e-3). Strong correlations emerged between global efficiency and token test (rs(46)=.52, p=1.2e-3). Node measures correlated significantly with AAT subtests, especially in the left hemisphere. The left precentral gyrus showed the strongest negative correlation between node strength and naming (rs(46)=-.51, p=1.5e-4), and betweenness centrality and token test (rs(46)=-.41, p=1.3e-2). The left superior frontal gyrus correlated most strongly with token test and betweenness centrality (rs(46)=.47, p=4.5e-3) and the left superior parietal gyrus correlated with token test and node strength (rs(46)=-.54, p=7.4e-4). Further, the left paracentral region correlated most strongly with naming and betweenness centrality (rs(46)=-.55, p=4.09e-4), speech comprehension and betweenness centrality (rs(46)=-.46, p=5.3e-3), and token test and betweenness centrality (rs(46)=-.67, p=1.1e-5). Network theory enables the mapping of impaired WM structure caused by tumors. Our analyses shed light on how a tumor affects a brain"s structural network and impairs the networks" integrity, leading to functional defects. Here, we identified significant sub-networks in relation to the gliomas" impact on the language network.

PD Dr. med.

Automated vs. nTMS-based analysis identification of the motor cortex and the corticospinal tract in motor eloquent brain tumors

Varia klinische Neurochirurgie

Automatisierte vs. nTMS-basierte Analyse zur Identifizierung des motorischen Kortex und des kortikospinalen Trakts bei motorisch eloquenten Hirntumoren

nTMS is a reliable tool for accurate identification of the motor cortex (M1) and corticospinal tract (CST) and allows prediction of individual risk for new/progressive postoperative paresis in motor-associated brain tumors. However, it requires appropriate technical and personnel resources. In this study, we assessed the feasibility of identifying M1 and CST based on an anatomical mapping-based algorithm of a navigation planning software in comparison to the clinically established and validated nTMS motor mapping.A retrospective analysis of a consecutive, prospective cohort of patients with brain tumors in motor eloquent areas who underwent preoperative nTMS motor mapping was performed. For both methods, the infiltration of M1 and CST, the tumor tract distance (TTD) and the fractional anisotropy (FA) were assessed. The agreement is measured with dice score and the overlap coefficient using the nTMS motor map as ground truth.101 patients (46 females) with a mean age of 51 years (range: 20-86 years) were included. All patients suffered from brain tumors (74 contrast-enhancing tumors, 29 tumor recurrences), of whom edema was present in M1 in 50 cases. The automated algorithm visualized the precentral gyrus in all but in one subject (99%) and the corticospinal tract in all subjects (100%). However, limited results with clearly implausible components (e.g. extended segmentations beyond M1 / restricted segmentations with only a few parts of M1 / tracts that clearly do not correspond to the course of the CST) were observed in 60 cases for M1 (59.4%) and 56 cases for CST (55.4%). Significant differences were found in TTD between the two models (mean+-SD: 4+-6 mm, p=.0001). On average, the overlap for M1 segmentation was 40% (range: 0-86%), and the dice score of the CST tractography was 35% (range: 6-78%).Automated anatomical-based segmentation of M1 and tractography of CST was feasible in almost all cases. However, nTMS is necessary for proper surgical planning, especially in cases where the tumor and perifocal edema interfere with the results of automated anatomical-based algorithm, highlighting the importance of individual functional mapping. Big data analysis and neural networks may help to improve the accuracy and efficiency of automated algorithms.

Graph network theory parameters reveal a high functional peritumoral network efficiency relevant for overall survival in treatment-naîve glioblastoma

Parameter der Graphentheorie weisen auf eine für das Gesamtüberleben relevante hohe funktionelle peritumorale Netzwerkeffizienz bei therapienaîven Glioblastomen hin.

When investigating tumor-related changes in the brain's functional connectome, graph network theory analyses allow promising insights into the dynamic functional interaction between a tumor and its peritumoral microenvironment. Regarding the peritumoral surrounding as a network, important information about its efficiency and its connectedness can be gained, allowing to estimate the tumor's impact on its immediate surrounding. We investigated whether network properties might be suitable candidates for the identification of peritumoral network efficiency and its relevance for overall survival (OS), considering surgical resection and adjuvant treatment.50 treatment-naîve glioblastoma patients (age: 65±9 years) were included and underwent resting-state fMRI. Tumor lesions were segmented and a 40mm tumor-surrounding peritumoral network was created. This peritumoral network was mirrored to the contralesional hemisphere, and lesional and contralesional peritumoral networks were analyzed by applying graph network theory techniques. Lesional and contralesional mean degree centrality (MDC) were compared, and the relevance for OS was analyzed. In a preliminary analysis, treatment-related changes in MDC differences were investigated in five patients with follow-up data.MDC was significantly higher in the lesional than in the contralesional peritumoral network (MDClesional=163, MDCcontra=137, p=.028), indicating a tumor-induced effect on its local surrounding, reflecting a high peritumoral network efficiency. Results further indicated a higher difference between lesional and contralesional MDC to be associated with shorter survival times (MDChigh=304 days, MDClow=581 days, p=.013). Treatment reduced the difference between lesional and contralesional MDC, an effect that was especially prominent in four patients with tumor resection and adjuvant therapy (MDCinitial=64, MDCfollow-up=16).A high difference between lesional and contralesional MDC suggested a high peritumoral network efficiency and reduced OS in treatment-naîve glioblastoma. This efficiency tended to decrease after tumor resection and adjuvant treatment, emphasizing the importance of surgery for the interruption of the functional communication between the tumor and its peritumoral surrounding. Investigating peritumoral network dynamics can aid in identifying network efficiency hubs, which can be addressed by individualized surgical treatment, thus improving oncological outcome.

Neuroplastic changes in the visual system and network connectivity in brain tumor patients with lesions in the optical radiation

Neuroplastische Veränderungen im visuellen System bei Patienten mit Hirntumor in der radiatio optica

The influence of brain tumors along important structures of the visual pathway on the overall organization of the visual system is not yet well understood. To address this, the influence of a brain tumor in the area of the optic radiation on the functional connectivity of the visual network was investigated in this study.fMRI data of 145 patients (76 male, 69 female, mean age 54.15 years) with brain tumors (96 left hemisphere, 49 right hemisphere) were included in this study. Based on the tumor location, patients were either assigned to a group with tumors in the area of the optic radiation (69 patients) or to a group that did not involve this structure (76 patients). A word generation task was chosen as fMRI paradigm, as this reliably addresses the visual system through the visual presentation of stimuli in written form. Based on these data, seed-to-voxel functional connectivity was calculated using the CONN toolbox. Here, four different Regions of Interest (ROIs) of the visual network were used as seed regions. These ROIs were located in the calcarine cortex of both hemispheres (CC), occipital poles of both hemispheres (OP), and lateral occipital cortex of each the left (LOCL) and right (LOCR) hemisphere.The results showed that patients with tumors distant from the optic radiation had higher connectivity regarding all four ROIs compared to patients with tumors located in the projection area of the optic radiation. Significant connectivity clusters included parts of the left planum temporale, Heschl"s gyrus, parietal operulum, and central opercular cortex for the CC (p = .013), OP (p < .001), and LOCR (p < .001) seed ROIs. Also, the left lateral occipital cortex was included in the significant connectivity clusters of the OP and LOCR seeds, while the same area of the right hemisphere was significant for the LOCL ROI (p = .045). To evaluate hemispheric effects, the patients were additionally subdivided according to the tumor-affected hemisphere. The results yieled similar significant connectivity profiles as the entire patient sample, except that the cluster sizes of the single significant connectivity clusters were reduced.This study showed differences between the two patient groups in functional connectivity within the visual network. The results suggest that not only primary visual areas but also higher visual areas appear to be affected, as significant voxels were not only found in the occipital lobe, but also in the parietal and temporal lobes.

Session 20
Tumor – Funktionelle Diagnostik

Termin

Vorsitz

Thema

Programm

Session 20Tumor – Funktionelle Diagnostik

Termin

Vorsitz

Thema

Programm

Session 20
Tumor – Funktionelle Diagnostik