Estimating Local Cellular Density in Glioma Using MR Imaging Data.

BACKGROUND AND PURPOSE: Increased cellular density is a hallmark of gliomas, both in the bulk of the tumor and in areas of tumor infiltration into surrounding brain. Altered cellular density causes altered imaging findings, but the degree to which cellular density can be quantitatively estimated from imaging is unknown. The purpose of this study was to discover the best MR imaging and processing techniques to make quantitative and spatially specific estimates of cellular density. MATERIALS AND METHODS: We collected stereotactic biopsies in a prospective imaging clinical trial targeting untreated patients with gliomas at our institution undergoing their first resection. The data included preoperative MR imaging with conventional anatomic, diffusion, perfusion, and permeability sequences and quantitative histopathology on biopsy samples. We then used multiple machine learning methodologies to estimate cellular density using local intensity information from the MR images and quantitative cellular density measurements at the biopsy coordinates as the criterion standard. RESULTS: The random forest methodology estimated cellular density with R 2 = 0.59 between predicted and observed values using 4 input imaging sequences chosen from our full set of imaging data (T2, fractional anisotropy, CBF, and area under the curve from permeability imaging). Limiting input to conventional MR images (T1 pre- and postcontrast, T2, and FLAIR) yielded slightly degraded performance (R2 = 0.52). Outputs were also reported as graphic maps. CONCLUSIONS: Cellular density can be estimated with moderate-to-strong correlations using MR imaging inputs. The random forest machine learning model provided the best estimates. These spatially specific estimates of cellular density will likely be useful in guiding both diagnosis and treatment.

Imaging-Based Algorithm for the Local Grading of Glioma.

BACKGROUND AND PURPOSE: Gliomas are highly heterogeneous tumors, and optimal treatment depends on identifying and locating the highest grade disease present. Imaging techniques for doing so are generally not validated against the histopathologic criterion standard. The purpose of this work was to estimate the local glioma grade using a machine learning model trained on preoperative image data and spatially specific tumor samples. The value of imaging in patients with brain tumor can be enhanced if pathologic data can be estimated from imaging input using predictive models. MATERIALS AND METHODS: Patients with gliomas were enrolled in a prospective clinical imaging trial between 2013 and 2016. MR imaging was performed with anatomic, diffusion, permeability, and perfusion sequences, followed by image-guided stereotactic biopsy before resection. An imaging description was developed for each biopsy, and multiclass machine learning models were built to predict the World Health Organization grade. Models were assessed on classification accuracy, Cohen κ, precision, and recall. RESULTS: Twenty-three patients (with 7/9/7 grade II/III/IV gliomas) had analyzable imaging-pathologic pairs, yielding 52 biopsy sites. The random forest method was the best algorithm tested. Tumor grade was predicted at 96% accuracy (κ = 0.93) using 4 inputs (T2, ADC, CBV, and transfer constant from dynamic contrast-enhanced imaging). By means of the conventional imaging only, the overall accuracy decreased (89% overall, κ = 0.79) and 43% of high-grade samples were misclassified as lower-grade disease. CONCLUSIONS: We found that local pathologic grade can be predicted with a high accuracy using clinical imaging data. Advanced imaging data improved this accuracy, adding value to conventional imaging. Confirmatory imaging trials are justified.

Performance Assessment for Brain MR Imaging Registration Methods.

BACKGROUND AND PURPOSE: Clinical brain MR imaging registration algorithms are often made available by commercial vendors without figures of merit. The purpose of this study was to suggest a rational performance comparison methodology for these products. MATERIALS AND METHODS: Twenty patients were imaged on clinical 3T scanners by using 4 sequences: T2-weighted, FLAIR, susceptibility-weighted angiography, and T1 postcontrast. Fiducial landmark sites (n = 1175) were specified throughout these image volumes to define identical anatomic locations across sequences. Multiple registration algorithms were applied by using the T2 sequence as a fixed reference. Euclidean error was calculated before and after each registration and compared with a criterion standard landmark registration. The Euclidean effectiveness ratio is the fraction of Euclidean error remaining after registration, and the statistical effectiveness ratio is similar, but accounts for dispersion and noise. RESULTS: Before registration, error values for FLAIR, susceptibility-weighted angiography, and T1 postcontrast were 2.07 ± 0.55 mm, 2.63 ± 0.62 mm, and 3.65 ± 2.00 mm, respectively. Postregistration, the best error values for FLAIR, susceptibility-weighted angiography, and T1 postcontrast were 1.55 ± 0.46 mm, 1.34 ± 0.23 mm, and 1.06 ± 0.16 mm, with Euclidean effectiveness ratio values of 0.493, 0.181, and 0.096 and statistical effectiveness ratio values of 0.573, 0.352, and 0.929 for rigid mutual information, affine mutual information, and a commercial GE registration, respectively. CONCLUSIONS: We demonstrate a method for comparing the performance of registration algorithms and suggest the Euclidean error, Euclidean effectiveness ratio, and statistical effectiveness ratio as performance metrics for clinical registration algorithms. These figures of merit allow registration algorithms to be rationally compared.

Design and initial evaluation of a treatment planning software system for MRI-guided laser ablation in the brain.

PURPOSE: An open-source software system for planning magnetic resonance (MR)-guided laser-induced thermal therapy (MRgLITT) in brain is presented. The system was designed to provide a streamlined and operator-friendly graphical user interface (GUI) for simulating and visualizing potential outcomes of various treatment scenarios to aid in decisions on treatment approach or feasibility. METHODS: A portable software module was developed on the 3D Slicer platform, an open-source medical imaging and visualization framework. The module introduces an interactive GUI for investigating different laser positions and power settings as well as the influence of patient-specific tissue properties for quickly creating and evaluating custom treatment options. It also provides a common treatment planning interface for use by both open-source and commercial finite element solvers. In this study, an open-source finite element solver for Pennes' bioheat equation is interfaced to the module to provide rapid 3D estimates of the steady-state temperature distribution and potential tissue damage in the presence of patient-specific tissue boundary conditions identified on segmented MR images. RESULTS: The total time to initialize and simulate an MRgLITT procedure using the GUI was [Formula: see text]5 min. Each independent simulation took [Formula: see text]30 s, including the time to visualize the results fused with the planning MRI. For demonstration purposes, a simulated steady-state isotherm contour [Formula: see text] was correlated with MR temperature imaging (N = 5). The mean Hausdorff distance between simulated and actual contours was 2.0 mm [Formula: see text], whereas the mean Dice similarity coefficient was 0.93 [Formula: see text]. CONCLUSIONS: We have designed, implemented, and conducted initial feasibility evaluations of a software tool for intuitive and rapid planning of MRgLITT in brain. The retrospective in vivo dataset presented herein illustrates the feasibility and potential of incorporating fast, image-based bioheat predictions into an interactive virtual planning environment for such procedures.