Tract profiles predict cognitive outomes associated with surgery in glioblastoma patients
Poster No:
2358
Submission Type:
Abstract Submission
Authors:
Yizhou Wan1,2, Ajay Halai1,3, Rohitashwa Sinha1,2, Haiyan Zheng1,4, Richard Mair1,2, Thomas Santarius1,2, Robert Morris1,2, Alexis Joannides1,2, Matt Lambon-Ralph1,3, Stephen Price1,2
Institutions:
1University of Cambridge, Cambridge, Cambridgeshire, 2Division of Neurosurgery, Department of Clinical Neurosciences, Cambridge, United Kingdom, 3MRC Cognition Brain Sciences Unit, Cambridge, United Kingdom, 4MRC Biostatistics Unit, Cambridge, United Kingdom
First Author:
Yizhou Wan, MB BS MPhil MRCS
University of Cambridge|Division of Neurosurgery, Department of Clinical Neurosciences
Cambridge, Cambridgeshire|Cambridge, United Kingdom
University of Cambridge|Division of Neurosurgery, Department of Clinical Neurosciences
Cambridge, Cambridgeshire|Cambridge, United Kingdom
Co-Author(s):
Ajay Halai, PhD
University of Cambridge|MRC Cognition Brain Sciences Unit
Cambridge, Cambridgeshire|Cambridge, United Kingdom
University of Cambridge|MRC Cognition Brain Sciences Unit
Cambridge, Cambridgeshire|Cambridge, United Kingdom
Rohitashwa Sinha, BMedSci(Hons) BMBS M
University of Cambridge|Division of Neurosurgery, Department of Clinical Neurosciences
Cambridge, Cambridgeshire|Cambridge, United Kingdom
University of Cambridge|Division of Neurosurgery, Department of Clinical Neurosciences
Cambridge, Cambridgeshire|Cambridge, United Kingdom
Haiyan Zheng, PhD
University of Cambridge|MRC Biostatistics Unit
Cambridge, Cambridgeshire|Cambridge, United Kingdom
University of Cambridge|MRC Biostatistics Unit
Cambridge, Cambridgeshire|Cambridge, United Kingdom
Richard Mair, MB ChB PhD FRCS (SN)
University of Cambridge|Division of Neurosurgery, Department of Clinical Neurosciences
Cambridge, Cambridgeshire|Cambridge, United Kingdom
University of Cambridge|Division of Neurosurgery, Department of Clinical Neurosciences
Cambridge, Cambridgeshire|Cambridge, United Kingdom
Thomas Santarius, MD PhD FRCS (SN)
University of Cambridge|Division of Neurosurgery, Department of Clinical Neurosciences
Cambridge, Cambridgeshire|Cambridge, United Kingdom
University of Cambridge|Division of Neurosurgery, Department of Clinical Neurosciences
Cambridge, Cambridgeshire|Cambridge, United Kingdom
Robert Morris, MB ChB FRCS (SN)
University of Cambridge|Division of Neurosurgery, Department of Clinical Neurosciences
Cambridge, Cambridgeshire|Cambridge, United Kingdom
University of Cambridge|Division of Neurosurgery, Department of Clinical Neurosciences
Cambridge, Cambridgeshire|Cambridge, United Kingdom
Alexis Joannides, MB BChir PhD FRCS (S
University of Cambridge|Division of Neurosurgery, Department of Clinical Neurosciences
Cambridge, Cambridgeshire|Cambridge, United Kingdom
University of Cambridge|Division of Neurosurgery, Department of Clinical Neurosciences
Cambridge, Cambridgeshire|Cambridge, United Kingdom
Matt Lambon-Ralph, PhD
University of Cambridge|MRC Cognition Brain Sciences Unit
Cambridge, Cambridgeshire|Cambridge, United Kingdom
University of Cambridge|MRC Cognition Brain Sciences Unit
Cambridge, Cambridgeshire|Cambridge, United Kingdom
Stephen Price
University of Cambridge|Division of Neurosurgery, Department of Clinical Neurosciences
Cambridge, Cambridgeshire|Cambridge, United Kingdom
University of Cambridge|Division of Neurosurgery, Department of Clinical Neurosciences
Cambridge, Cambridgeshire|Cambridge, United Kingdom
Introduction:
Glioblastoma is the most common and aggressive primary brain tumour. They are diffusely infiltrative and invade white-matter tracts. (Price et al. 2017) Cognitive deficits are common, affecting over 90% of patients. (Sinha, Stephenson, and Price 2019) The effect of the tumour and surgery on cognition is poorly understood. Tractometry has been used predict motor and language outcomes in glioma patients. (Shams et al. 2022; Prasse et al. 2022)
We hypothesize that tractometry can be used to predict postoperative cognitive outcomes in glioblastoma patients by correlating microstructural measures from tract profiles with cognitive outcomes in a prospective cohort of glioblastoma patients undergoing surgery.
We hypothesize that tractometry can be used to predict postoperative cognitive outcomes in glioblastoma patients by correlating microstructural measures from tract profiles with cognitive outcomes in a prospective cohort of glioblastoma patients undergoing surgery.
Methods:
40 patients were tested pre-surgery (t0) and post-surgery using the OCS-Bridge Cognitive Screening Tool (t2). Patients were scanned with anatomical and multi-shell diffusion sequences preoperatively and postoperatively. Cognitive scores at each timepoint were z-scored based the preoperative mean and standard-deviation. PCA was used to identify the latent dimensions.
Anatomical processing involved tumour segmentation and lesion-filling. (Radwan et al. 2021) Whole-brain tractography was performed using Tractoflow. Diffusion tensor, Kurtosis, NODDI and RISH microstructural models were fitted. Tract segmentation was performed using TractSeg. (Wasserthal, Neher, and Maier-Hein 2018) Profiles were generated with 100 nodes per tract for 72 tracts in TractSeg.
Longitudinal tract segmentations were used to calculate bundle reproducibility based on weighted-DICE. (Edde et al. 2023) Reproducible tracts (weighted DICE > 0.4) were selected for downstream analysis. Sparse group lasso regression (SGLR) was used to predict t0 and t2 cognitive scores using preoperative profiles. (Richie-Halford et al. 2021) Model performance was measured using cross-validated mean squared error between predicted and observed scores. (Halai, Woollams, and Lambon Ralph 2020)
Anatomical processing involved tumour segmentation and lesion-filling. (Radwan et al. 2021) Whole-brain tractography was performed using Tractoflow. Diffusion tensor, Kurtosis, NODDI and RISH microstructural models were fitted. Tract segmentation was performed using TractSeg. (Wasserthal, Neher, and Maier-Hein 2018) Profiles were generated with 100 nodes per tract for 72 tracts in TractSeg.
Longitudinal tract segmentations were used to calculate bundle reproducibility based on weighted-DICE. (Edde et al. 2023) Reproducible tracts (weighted DICE > 0.4) were selected for downstream analysis. Sparse group lasso regression (SGLR) was used to predict t0 and t2 cognitive scores using preoperative profiles. (Richie-Halford et al. 2021) Model performance was measured using cross-validated mean squared error between predicted and observed scores. (Halai, Woollams, and Lambon Ralph 2020)
Results:
Attention and executive function accounted for 65% of the variance for OCS while perception and attention/visual memory accounted for 51% of the variance of the Cambridge perception screen.
Overall weighted-DICE was low (range: 0 - 0.38, µ: 0.22). Reproducibility was higher for tracts contralateral to the tumour in resection compared to biopsy patients (n = 6) for both right-sided (0.302 vs. 0.235 FWE-corrected p = 0.02) (Fig 1) and left-sided tumours (0.468 vs 0.254, FWE-corrected p < 0.01) (Fig 2). 23 tracts were selected for statistical modelling. Microstructural features loaded onto two components representing measures of tissue "hindrance" and "complexity" accounting for 68% of the variance.
Hindrance profiles predicted t0 and t2 scores. For t0 scores. K-fold cross-validated (k = 10) R2 were; attention (0.424, p = 5.45x10-6), executive function (0.294, p = 3.01-4), perception (0.177, 6.86x10-3), visual memory (0.229, p = 1.77x10--3). For t2 scores, the R2 values ranged between 0.154-0.565. Complexity explained higher variability in t2 scores (R2 range: 0.397 - 0.799) but not t0 scores.
Overall weighted-DICE was low (range: 0 - 0.38, µ: 0.22). Reproducibility was higher for tracts contralateral to the tumour in resection compared to biopsy patients (n = 6) for both right-sided (0.302 vs. 0.235 FWE-corrected p = 0.02) (Fig 1) and left-sided tumours (0.468 vs 0.254, FWE-corrected p < 0.01) (Fig 2). 23 tracts were selected for statistical modelling. Microstructural features loaded onto two components representing measures of tissue "hindrance" and "complexity" accounting for 68% of the variance.
Hindrance profiles predicted t0 and t2 scores. For t0 scores. K-fold cross-validated (k = 10) R2 were; attention (0.424, p = 5.45x10-6), executive function (0.294, p = 3.01-4), perception (0.177, 6.86x10-3), visual memory (0.229, p = 1.77x10--3). For t2 scores, the R2 values ranged between 0.154-0.565. Complexity explained higher variability in t2 scores (R2 range: 0.397 - 0.799) but not t0 scores.
·Weighted DICE for contralateral tracts (Right-sided tumours))
·Weighted DICE for contralateral tracts (Left-sided tumours))
Conclusions:
Tractometry-derived profiles describing microstructural features of preoperative white-matter tracts may predict postoperative cognition, offering prognostic information for glioblastoma patients. A multivariable model accounting for features grouped into white-matter tracts show similar performance when predicting preoperative and postoperative outcomes. This suggests that postoperative cognition may be related to preoperative measures of white-matter damage related to brain-intrinsic or tumour effects. Features representing signal orientational-dispersion which may reflect local structural complexity shows higher correlations with postoperative scores. Future work should aim to investigate how anatomical changes related to surgery correlate with tract reproducibility and to validate these models using held-out datasets.
Higher Cognitive Functions:
Executive Function, Cognitive Control and Decision Making
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural)
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
White Matter Anatomy, Fiber Pathways and Connectivity 2
Novel Imaging Acquisition Methods:
Diffusion MRI 1
Perception, Attention and Motor Behavior:
Attention: Visual
Keywords:
Cognition
Data analysis
Modeling
Multivariate
Statistical Methods
White Matter
WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC
Other - Brain tumour
1|2Indicates the priority used for review
Provide references using author date format
Caron, Guillaume Gilbert, Jean‐Christophe Houde, et al. 2023. ‘High‐frequency Longitudinal White Matter Diffusion‐ and Myelin‐based MRI Database: Reliability and Variability’. Human Brain Mapping 44 (9): 3758–80.
Halai, Ajay D., Anna M. Woollams, and Matthew A. Lambon Ralph. 2020. ‘Investigating the Effect of Changing Parameters When Building Prediction Models for Post-Stroke Aphasia’. Nature Human Behaviour 4 (7): 725–35.
Prasse, Gordian, Hans-Jonas Meyer, Cordula Scherlach, Jens Maybaum, Anastasia Hoffmann, Johannes Kasper, Michael Karl Fehrenbach, et al. 2022. ‘Preoperative Language Tract Integrity Is a Limiting Factor in Recovery from Aphasia after Glioma Surgery’. NeuroImage :
Clinical 37 (December): 103310.
Price, Stephen J., Kieren Allinson, Hongxiang Liu, Natalie R. Boonzaier, Jiun-Lin Yan, Victoria C. Lupson, and Timothy J. Larkin. 2017. ‘Less Invasive Phenotype Found in Isocitrate Dehydrogenase–Mutated Glioblastomas than in Isocitrate Dehydrogenase Wild-Type
Glioblastomas: A Diffusion-Tensor Imaging Study’. Radiology 283 (1)
Radwan, Ahmed M., Louise Emsell, Jeroen Blommaert, Andrey Zhylka, Silvia Kovacs, Tom Theys, Nico Sollmann, Patrick Dupont, and Stefan Sunaert. 2021. ‘Virtual Brain Grafting: Enabling Whole Brain Parcellation in the Presence of Large Lesions’. NeuroImage 229 (April): 117731.
Richie-Halford, Adam, Jason D. Yeatman, Noah Simon, and Ariel Rokem. 2021. ‘Multidimensional Analysis and Detection of Informative Features in Human Brain White Matter’. PLOS Computational Biology 17 (6): e1009136.
Shams, Boshra, Ziqian Wang, Timo Roine, Dogu Baran Aydogan, Peter Vajkoczy, Christoph Lippert, Thomas Picht, and Lucius S. Fekonja. 2022. ‘Machine Learning-Based Prediction of Motor Status in Glioma Patients Using Diffusion MRI Metrics along the Corticospinal Tract’. Brain Communications 4 (3): fcac141.
Sinha, Rohitashwa, Jade Marie Stephenson, and Stephen John Price. 2019. ‘A Systematic Review of Cognitive Function in Patients with Glioblastoma Undergoing Surgery’. Neuro Oncology Practice.
Wasserthal, Jakob, Peter Neher, and Klaus H. Maier-Hein. 2018. ‘TractSeg - Fast and Accurate White Matter Tract Segmentation’. NeuroImage 183 (December): 239–53.