Diffusion histology imaging differentiates distinct pediatric brain tumor histology

Abstract

High-grade pediatric brain tumors exhibit the highest cancer mortality rates in children. While conventional MRI has been widely adopted for examining pediatric high-grade brain tumors clinically, accurate neuroimaging detection and differentiation of tumor histopathology for improved diagnosis, surgical planning, and treatment evaluation, remains an unmet need in their clinical management. We employed a novel Diffusion Histology Imaging (DHI) approach employing diffusion basis spectrum imaging (DBSI) derived metrics as the input classifiers for deep neural network analysis. DHI aims to detect, differentiate, and quantify heterogeneous areas in pediatric high-grade brain tumors, which include normal white matter (WM), densely cellular tumor, less densely cellular tumor, infiltrating edge, necrosis, and hemorrhage. Distinct diffusion metric combination would thus indicate the unique distributions of each distinct tumor histology features. DHI, by incorporating DBSI metrics and the deep neural network algorithm, classified pediatric tumor histology with an overall accuracy of 85.8%. Receiver operating analysis (ROC) analysis suggested DHI's great capability in distinguishing individual tumor histology with AUC values (95% CI) of 0.984 (0.982-0.986), 0.960 (0.956-0.963), 0.991 (0.990-0.993), 0.950 (0.944-0.956), 0.977 (0.973-0.981) and 0.976 (0.972-0.979) for normal WM, densely cellular tumor, less densely cellular tumor, infiltrating edge, necrosis and hemorrhage, respectively. Our results suggest that DBSI-DNN, or DHI, accurately characterized and classified multiple tumor histologic features in pediatric high-grade brain tumors. If these results could be further validated in patients, the novel DHI might emerge as a favorable alternative to the current neuroimaging techniques to better guide biopsy and resection as well as monitor therapeutic response in patients with high-grade brain tumors.

Figure 1

Illustration of brain specimen procurement from a patient with high-grade pediatric brain tumor. (a) In vivo Gd-enhanced T1-weighted image indicated a large lesion (square) with heterogeneous intensities in the right posterior region from a 16-year-old patient with embryonal neoplasm (WHO Grade IV). (b) Brain specimen was procured and immediately formalin-fixed. (c) Coronal slices revealed a large tumor with admixed hemorrhage and necrosis in the right thalamus (arrow). (d) Five tissue blocks were prepared in total i.e. from tumor (block 2, blcok 4), tumor interface with normal adjacent brain (block 3), hemorrhage and necrosis (block 5), as well as grossly normal brain (block 1) (c).

Figure 2

Co-registration between histology and MRI. Raw RGB H&E images were first converted to grayscale images for enhanced coregistration. Eighteen pairs of landmarks along the perimeter of the brain specimens were manually placed on the MR image and grayscale H&E image. The transformation matrix of the two-dimensional thin plate spline (TPS) registration was computed in MIPAV (version 10.0.0) and applied to warp the H&E image to the orientation of the MR image. After co-registration, the pathologist defined tumor histology regions on H&E images. These were then successfully transferred to the corresponding MR image.

Figure 3

A representative tissue block was imaged with ex vivo MRI, followed by histologic processing and evaluation. (a) H&E image of a sectioned specimen after ex vivo MRI with regions of white matter, densely cellular tumor and hemorrhage outlined for assessing the efficacy of metrics derived by multi-parametric MRI and DBSI. The expanded region of densely cellular tumor features characteristically increased cellularity. Scale bar measures 50 um. (b) T1W and T2W MRI did not distinguish densely cellular tumor region from white matter or hemorrhage. Most strikingly, diffusely cellular tumor region exhibited lower DWI and higher ADC countering the conventional wisdom that higher tumor cellularity is associated with restricted diffusion. WM white matter, DC tumor densely cellular tumor.

Figure 4

Group analysis on different tumor histologic components on representative diffusion metrics including (a) ADC, (b) DTI FA, (c) DBSI isotropic ADC, (d) highly restricted fraction, (e) restricted fraction, (f) hindered fraction, (g) free fraction, and (h) fiber fraction. Particularly, normal WM and the infiltrative edge showed higher fiber fraction and DTI-FA than the other tumor histologies. DC tumor and LDC tumor showed higher restricted fraction values than other histologies. Necrosis showed higher ADC, hindered fraction and free fraction values as well as lower restricted fraction, fiber fraction and DTI-FA compared to the other histologies. These findings were collectively consistent with DBSI’s modelling for malignant brain tumor. ADC, µm²/ms. Normal WM normal white matter, DC tumor densely cellular tumor, LDC tumor less densely cellular tumor.

Figure 5

(a) Representative H&E images of normal white matter, densely cellular tumor, less densely cellular tumor, infiltrative edge, necrosis and hemorrhage, respectively. (b) Independent test dataset confusion matrix for the predictions of DHI versus gold standard, i.e. histologic examination (n = 9939). Rows contain tumor histologic classifications identified by a neuropathologist, and columns represent tumor histologic classifications as predicted by DHI. Scale bar measures 100 μm. Normal WM normal white matter, DC tumor densely cellular tumor, LDC tumor less densely cellular tumor.

Figure 6

Receiver operating characteristics (ROC) curves and precision-recall (PR) curves calculated using one-vs-rest strategy for 6 different tumor histological components including (a) normal white matter, (b) densely cellular tumor, (c) less densely cellular tumor, (d) tumor infiltrative edge, (e) tumor necrosis and (f) hemorrhage. All 6 ROC curves showed high areas under curve (AUC), indicating strong sensitivity and specificity in detecting these tumor histologic components. Tumor infiltrative edge did not perform as well as other histologic components in precision-recall analysis, indicating that tumor infiltration could be overestimated by the model. Normal WM normal white matter, DC tumor densely cellular tumor, LDC tumor less densely cellular tumor.