Glioblastoma (GBM) is a heterogeneous and lethal brain cancer. These tumors are followed using magnetic resonance imaging (MRI), which is unable to precisely identify tumor cell invasion, impairing effective surgery and radiation planning. We present a novel hybrid model, based on multiparametric intensities, which combines machine learning (ML) with a mechanistic model of tumor growth to provide spatially resolved tumor cell density predictions. The ML component is an imaging data-driven graph-based semi-supervised learning model and we use the Proliferation-Invasion (PI) mechanistic tumor growth model. We thus refer to the hybrid model as the ML-PI model. The hybrid model was trained using 82 image-localized biopsies from 18 primary GBM patients with pre-operative MRI using a leave-one-patient-out cross validation framework. A Relief algorithm was developed to quantify relative contributions from the data sources. The ML-PI model statistically significantly outperformed (p < 0.001) both individual models, ML and PI, achieving a mean absolute predicted error (MAPE) of 0.106 ± 0.125 versus 0.199 ± 0.186 (ML) and 0.227 ± 0.215 (PI), respectively. Associated Pearson correlation coefficients for ML-PI, ML, and PI were 0.838, 0.518, and 0.437, respectively. The Relief algorithm showed the PI model had the greatest contribution to the result, emphasizing the importance of the hybrid model in achieving the high accuracy.

BACKGROUND: MRI-based modeling of tumor cell density (TCD) can significantly improve targeted treatment of Glioblastoma (GBM). Unfortunately, interpatient variability limits the predictive ability of many modeling approaches. We present a Transfer Learning (TL) method that generates individualized patient models, grounded in the wealth of population data, while also detecting and adjusting for interpatient variabilities based on each patient’s own histologic data. METHODS: We recruited primary GBM patients undergoing image-guided biopsies and preoperative imaging including contrast-enhanced MRI (CE-MRI), Dynamic-Susceptibility-Contrast (DSC)-MRI, and Diffusion Tensor Imaging (DTI). We calculated relative cerebral blood volume (rCBV) from DSC-MRI and mean diffusivity (MD) and fractional anisotropy (FA) from DTI. Following image coregistration, we assessed TCD for each biopsy and identified corresponding localized MRI measurements. We then explored a range of univariate and multivariate predictive models of TCD based on MRI measurements in a generalized one-model-fits-all (OMFA) approach. We then implemented both univariate and multivariate individualized TL predictive models, which harness the available population level data but allow for individual variability in their predictions. Finally, we compared Pearson correlation coefficients and mean absolute error between the individualized TL versus generalized OMFA models. RESULTS: TCD significantly correlated with rCBV (r=0.33,p<0.0001) and T1+C (r=0.36,p<0.0001) on univariate analysis after correcting for multiple comparisons. With single variable modeling (using rCBV), TL increased predictive performance (r=0.53, MAE=15.19%) compared to OMFA (r=0.27, MAE=17.79%). With multivariate modeling, TL further improved performance (r=0.88, MAE 5.66%) compared to OMFA (r=0.39, MAE=16.55%). CONCLUSION: TL significantly improves predictive modeling performance for quantifying tumor cell density in GBM.

Background: Alzheimer's disease (AD) is a progressive neurodegenerative disease that affects aging populations. Current MRI techniques are often limited in their sensitivity to underlying neuropathological changes. Purpose: To characterize differences in voxel-based morphometry (VBM), apparent diffusion coefficient (ADC), and intravoxel incoherent motion diffusion-weighted imaging (IVIM-DWI) metrics in aging populations. Additionally, to investigate the connection between cognitive assessments and neuroimaging metrics. Study Type: Prospective/cross-sectional. Population: In all, 49 subjects, including 13 with AD dementia, 12 with mild cognitive impairment (MCI), and 24 healthy controls (HC). Field Strength/Sequence: 3T/magnetization-prepared rapid acquisition gradient echo (MP-RAGE) and IVIM-DWI Assessment: All participants completed a cognitive screening battery prior to MRI. IVIM-DWI maps (pure diffusion coefficient [D], pseudodiffusion coefficient [D*], and perfusion fraction [f]) were generated from a biexponential fit of diffusion MRI data. VBM was performed on the standard T 1-weighted MP-RAGE structural images. Group-wise templates were used to compare across groups. Statistical Tests: Analysis of covariance (ANCOVA) with gender and age as covariates (familywise error [FWE] corrected, post-hoc comparisons using Bonferroni correction) for group comparisons. Partial-η 2 and Hedges' g were used for effectsize analysis. Spearman's correlations (false discovery rate [FDR]-corrected) for the relationship between cognitive scores and imaging. Results: Clusters of significant group-wise differences were found mainly in the temporal lobe, hippocampus, and amygdala using all VBM and IVIM methods (P < 0.05 FWE). While VBM showed significant changes between MCI and AD groups and between HC and AD groups, no significant clusters were observed between HC and MCI using VBM. ADC and IVIM-D demonstrated significant changes, at P < 0.05 FWE, between HC and MCI, notably in the amygdala and hippocampus. Several voxel-based correlations were observed between neuroimaging metrics and cognitive tests within the cognitively impaired groups (P < 0.05 FDR). Data Conclusion: These findings suggest that IVIM-DWI metrics may be earlier biomarkers for AD-related changes than VBM. The use of these techniques may provide novel insight into subvoxel neurodegenerative processes. Level of Evidence: 2 Technical Efficacy Stage: 2

Background:The integrity and connectivity of the frontal lobe, which subserves fluency, may be compromised by both ASD and aging. Alternate networks often integrate to help compensate for compromised functions during aging. We used network analyses to study how compensation may overcome age-related compromised in individuals with ASD.Method: Participants consisted of middle-aged (40-60; n=24) or young (18-25; n=18) righthanded males who have a diagnosis of ASD, and age-and IQ-matched control participants (n=20, 14, respectively). All performed tests of language and executive functioning and a fluency functional MRI task. We first used group individual component analysis (ICA) for each of the 4 groups to determine whether different networks were engaged. An SPM analysis was used to compare activity detected in the network nodes from the ICA analyses. Results:The individuals with ASD performed more slowly on two cognitive tasks (Stroop word reading and Trailmaking Part A). The 4 groups engaged different networks during the fluency fMRI task despite equivalent performance. Comparisons of specific regions within these networks indicated younger individuals had greater engagement of the thalamus and supplementary speech area, while older adults engaged the superior temporal gyrus. Individuals with ASD did not disengage from the Default Mode Network during word generation. Conclusion:Interactions between diagnosis and aging were not found in this study of young and middle-aged men, but evidence for differential engagement of compensatory networks was observed.

Purpose Brain tumor dynamic susceptibility contrast (DSC) MRI is adversely impacted by T1 and T2∗ contrast agent leakage effects that result in inaccurate hemodynamic metrics. While multi‐echo acquisitions remove T1 leakage effects, there is no consensus on the optimal set of acquisition parameters. Using a computational approach, we systematically evaluated a wide range of acquisition strategies to determine the optimal multi‐echo DSC‐MRI perfusion protocol. Methods Using a population‐based DSC‐MRI digital reference object (DRO), we assessed the influence of preload dosing (no preload and full dose preload), field strength (1.5 and 3T), pulse sequence parameters (echo time, repetition time, and flip angle), and leakage correction on relative cerebral blood volume (rCBV) and flow (rCBF) accuracy. We also compared multi‐echo DSC‐MRI protocols with standard single‐echo protocols. Results Multi‐echo DSC‐MRI is highly consistent across all protocols, and multi‐echo rCBV (with or without use of a preload dose) had higher accuracy than single‐echo rCBV. Regression analysis showed that choice of repetition time and flip angle had minimal impact on multi‐echo rCBV and rCBV, indicating the potential for significant flexibility in acquisition parameters. The echo time combination had minimal impact on rCBV, though longer echo times should be avoided, particularly at higher field strengths. Leakage correction improved rCBV accuracy in all cases. Multi‐echo rCBF was less biased than single‐echo rCBF, although rCBF accuracy was reduced overall relative to rCBV. Conclusions Multi‐echo acquisitions were more robust than single‐echo, essentially decoupling both repetition time and flip angle from rCBV accuracy. Multi‐echo acquisitions obviate the need for preload dosing, although leakage correction to remove residual T2∗ leakage effects remains compulsory for high rCBV accuracy.