Search

INTRODUCTION The quantification of intratumoral heterogeneity – through radiomics-based approaches - can help resolve the regionally distinct genetic drug targets that may co-exist within a single Glioblastoma (GBM) tumor. While this offers potential diagnostic value under the paradigm of individualized oncology, clinical decision-making must also consider the degree of uncertainty associated with each model. In this study, we evaluate the performance of a novel machine-learning (ML) algorithm, called Gaussian Process (GP) modeling, that can quantify the impact of multiple sources of uncertainty in ML model development and prediction accuracy, including variabilities in the copy number measurement, radiomics features, training sample characteristics, and training sample size. METHOD We collected 95 image-localized biopsies from 25 primary GBM patients. We coregistered stereotactic locations with preoperative multi-parametric MRI features (conventional MRI, DSC perfusion, Diffusion Tensor Imaging) to generate spatially matched pairs of MRI and copy number variants (CNV) for for each biopsy. We developed a Gaussian Process (GP) model to predict CNV for Epidermal Growth Factor Receptor (EGFR) based on MRI radiomic features in each patient. We used leave-one-patient-out cross validation to quantify prediction accuracy and model uncertainty. Spatial prediction and uncertainty (p-value) maps were overlaid to help visualize regional genetic variation of EGFR and uncertainty of the radiomic predictions. RESULT: The initial GP radiomics model for EGFR amplification (CNV > 3.5) produced a sensitivity of 0.8 and specificity of 0.8. Samples/regions associated with high uncertainty (p-value >0.05) correlated with either 1) extrapolation of radiomic features from the training set-defined feature space or 2) insufficient training samples in the feature space. CONCLUSION We present a ML-based model that quantifies spatial genetic heterogeneity in GBM, while also estimating model uncertainties that result from multi-source data variabilities. This approach lays the groundwork for prospective clinical integration of modeling-based diagnostic approaches in the paradigm of individualized medicine.

INTRODUCTION Neuroimaging dogma for glioblastoma asserts that hyperintensity on T1Gd MRI reveals the bulk of the tumor, while T2/FLAIR signal indicates edema. However, it is unclear whether this edema results from immune response or increased tumor cells. Further, one significant driver of the known sex differences in glioblastoma may be differences in immune response, due to the X-linkage of many immune genes. Based on this, we hypothesized that assumptions regarding tumor cellularity in T2/FLAIR images should be tailored to the biological sex of the patient. METHODS Using a retrospective cohort of 18 primary glioblastoma patients receiving multiple image-localized biopsies (82 total) and standard MRI, we assessed: distance of biopsy from T1Gd and T2 areas; a pathologist’s score of percent tumor cell density; and an imaging-based invasion metric, D/ρ. This metric is derived from the biomathematical Proliferation-Invasion model of glioma growth, which features two parameters, net growth rate (ρ) and net invasion rate (D). Their ratio D/ρ is related to degree of invasion, and is estimated from volumetric measurements of MRI abnormalities. Additionally, 25 patient-derived xenograft models implanted in females were grown until moribund, at which point brains were excised and stained for DAPI (to show all cells) and Lamin (to highlight tumor cells). Image processing of lamin-stained sections defines contours of intensity correlating with cell density. RESULTS Outside both the T1Gd and T2 region, male patient biopsies had higher tumor cell densities than females. Males also tended to have higher invasion metrics. Although each set derived from different patients, preclinical metrics of invasion were positively correlated with clinical invasion in females but negatively correlated in males. CONCLUSION Our preliminary finding that cell distribution patterns correlate with imaging metrics differently between the sexes supports the hypothesis that the degree of tumor cell density represented on certain MRI sequences may be sex-specific.

Dynamic susceptibility contrast (DSC) MRI measures of brain tumor cerebral blood volume (CBV) are able to predict grade, overall survival and response to treatment. Wide-spread acceptance of DSC-MRI has been challenged by the need to balance contrast agent dose and CBV accuracy. The goal of this study was to identify and validate single-dose, BTIP compliant, DSC-MRI protocols. Using a validated, patient-based DRO, we evaluated CBV accuracy across a range of acquisition parameters (field strength, TR, TE, flip angle, multi-echo acquisitions, dosing protocols) and post-processing steps. To validate the optimal protocols, we next collected DSC-MRI data following ASFNR’s recommended “double – dose” approach, where a single-dose preload (to minimize T1 effects) is given prior to a second bolus injection (for DSC-MRI data acquisition). The single-dose DSC-MRI data was collected during the preload bolus injection. Consistency of the derived CBV data, visual agreement and data characteristics (e.g. CNR) was statistically evaluated. When using a single-dose and routine single-echo pulse sequence, the DRO analysis found that a low flip angle (LFA = 30o) and 30ms TE provided the highest CBV accuracy (concordance correlation coefficient (CCC) = 0.92) and precision (coefficient of variation (CV) = 8.2%)). For comparison, the maximum accuracy found with the DRO utilizes a double-dose injection protocol and yielded a CCC of 0.98 and CV of 6.8%. Single-dose, multi-echo acquisitions provided higher accuracy than the LFA data and matched that found with the double-dose approach. In patients (data collection ongoing), the agreement between single-dose LFA (n > 40) or multi-echo (n > 40) based CBV values and the reference double-dose approach was very high (CCC > 0.94) and were statistically equivalent. Optimized single-dose DSC-MRI protocols provide highly accurate CBV data, use lower doses of contrast agent, and simplify scan procedures, indicating their potential for robust use in clinical practice and trials.

BACKGROUND An important challenge in radiomics research is reproducibility. Images are collected on different image scanners and protocols, which introduces significant variability even for the same type of image across institutions. In the present proof-of-concept study, we address the reproducibility issue by using domain adaptation – an algorithm that transforms the radiomic features of each new patient to align with the distribution of features formed by the patient samples in a training set. METHOD Our dataset included 18 patients in training with a total of 82 biopsy sample. The pathological tumor cell density was available for each sample. Radiomic (statistical + texture) features were extracted from the region of six image contrasts locally matched with each biopsy sample. A Gaussian Process (GP) classifier was built to predict tumor cell density using radiomic features. Another 6 patients were used to test the training model. These patients had a total of 31 biopsy samples. The images of each test patient were purposely normalized using a different approach, i.e., using the CSF instead of the whole brain as the reference. This was to mimic the practical scenario of image source discrepancy between different patients. Domain adaptation was applied to each test patient. RESULTS Among the 18 training patients, the leave-one-patient-out cross validation accuracy is 0.81 AUC, 0.78 sensitivity, and 0.83 specificity. When the trained model was applied to the 6 test patients (purposely normalized using a different approach than that of the training data), the accuracy dramatically reduced to 0.39 AUC, 0.08 sensitivity, and 0.61 specificity. After using domain adaption, the accuracy improved to 0.68 AUC, 0.62 sensitivity, and 0.72 specificity. CONCLUSION We provide candidate enabling tools to address reproducibility in radiomics models by using domain adaption algorithms to account for discrepancy of the images between different patients.