An accessible deep learning tool for voxel-wise classification of brain malignancies from perfusion MRI

Noninvasive differential diagnosis of brain tumors is currently based on the assessment of magnetic resonance imaging (MRI) coupled with dynamic susceptibility contrast (DSC). However, a definitive diagnosis often requires neurosurgical interventions that compromise patients' quality of life. We apply deep learning on DSC images from histology-confirmed patients with glioblastoma, metastasis, or lymphoma. The convolutional neural network trained on ∼50,000 voxels from 40 patients provides intratumor probability maps that yield clinical-grade diagnosis. Performance is tested in 400 additional cases and an external validation cohort of 128 patients. The tool reaches a three-way accuracy of 0.78, superior to the conventional MRI metrics cerebral blood volume (0.55) and percentage of signal recovery (0.59), showing high value as a support diagnostic tool. Our open-access software, Diagnosis In Susceptibility Contrast Enhancing Regions for Neuro-oncology (DISCERN), demonstrates its potential in aiding medical decisions for brain tumor diagnosis using standard-of-care MRI.

Differential diagnosis of central lymphoma and high-grade glioma: dynamic contrast-enhanced histogram

BACKGROUND

In clinical diagnosis, some central nervous system lymphomas (CNSL) are difficult to distinguish from high-grade gliomas (HGG).

PURPOSE

To evaluate the diagnostic efficacy of the histogram analysis of dynamic contrast-enhanced (DCE) magnetic resonance imaging (MRI) in the identification of CNSL and HGG.

MATERIAL AND METHODS

In all, 43 patients diagnosed with HGG (n = 28) and CNSL (n = 15) by histopathology underwent DCE-MRI scanning. Differences in histogram parameters based on DCE-MRI between HGG and CNSL were analyzed by Mann-Whitney U test. In addition, receiver operating characteristic (ROC) analysis was performed. Short-term follow-up of patients was performed using Kaplan-Meier analysis to explore the survival rates of HGG and CNSL.

RESULTS

For the ROC curve analysis, we demonstrate that the 10th percentile of K^trans (area under the curve [AUC] = 0.912, sensitivity = 86.7%, specificity = 92.9%), K_ep (AUC = 0.940, sensitivity = 93.3%, specificity = 79.6%), V_e (AUC = 0.907, sensitivity = 86.7%, specificity = 89.3%), and AUC (AUC = 0.904, sensitivity = 86.7%, specificity = 92.9%) were significantly different between the CNSL and HGG groups (P < 0.001), with high diagnostic efficiency. Table 2 shows that the histogram features based on AUC maps (10th, 25th, median, 75th, 90th, and mean) were always significantly higher in the CNSL group than in the HGG group (P < 0.001). There was no significant difference in V_p or in the 75th, 90th and mean of K^trans, K_ep, and Ve between the CNSL and HGG groups (P > 0.05).

CONCLUSION

A histogram analysis of DCE-MRI identified significant differences between HGG and CNSL, and this will help in the clinical differential diagnosis of these conditions.

Correlation and Characteristics of Intravoxel Incoherent Motion and Arterial Spin Labeling Techniques Versus Multiple Parameters Obtained on Dynamic Susceptibility Contrast Perfusion MRI for Brain Tumors

Purpose : To compare data on brain tumors derived from intravoxel incoherent motion (IVIM) and arterial spin labeling (ASL) imaging with multiple parameters obtained on dynamic susceptibility contrast (DSC) perfusion MRI and to clarify the characteristics of IVIM and ASL perfusion data from the viewpoint of cerebral blood flow (CBF) analysis. Methods : ASL-CBF and IVIM techniques as well as DSC examination were performed in 24 patients with brain tumors. The IVIM data were analyzed with the two models. The relative blood flow (rBF), relative blood volume (rBV) corrected relative blood volume (crBV), mean transit time (MTT), and leakage coefficient (K2) were obtained from the DSC MRI data. Results : The ASL-CBF had the same tendency as the perfusion parameters derived from the DSC data, but the permeability from the vessels had less of an effect on the ASL-CBF. The diffusion coefficient of the fast component on IVIM contained more information on permeability than the f value. Conclusion : ASL-CBF is more suitable for the evaluation of perfusion in brain tumors than IVIM parameters. ASL-CBF and IVIM techniques should be carefully selected and the biological significance of each parameter should be understood for the correct comprehension of the pathological status of brain tumors. J. Med. Invest. 66 : 308-313, August, 2019.

Correlation of Tumor Perfusion Between Carbon-13 Imaging with Hyperpolarized Pyruvate and Dynamic Susceptibility Contrast MRI in Pre-Clinical Model of Glioblastoma

PURPOSE

The purpose of this study was to compare C-13 imaging parameters with hyperpolarized [1-¹³C]pyruvate with conventional gadolinium (Gd)-based perfusion weighted imaging using an orthotopic xenograft model of human glioblastoma multiforme (GBM).

PROCEDURES

C-13 3D magnetic resonance spectroscopic imaging (MRSI) data were obtained from 14 tumor-bearing rats after the injection of hyperpolarized [1-¹³C]pyruvate at a 3T scanner. Dynamic susceptibility contrast (DSC) perfusion-weighted MR images were obtained following intravenous administration of Gd-DTPA. Normalized lactate, pyruvate, total carbon, and lactate to pyruvate ratio from C-13 MRSI data were compared with normalized peak height and percent recovery of ΔR2* curve from the DSC images in the voxels containing tumor using a Pearson's linear correlation.

RESULTS

Normalized peak height from DSC imaging showed substantial correlations with normalized lactate (r = 0.6, p = 0.02) and total carbon (r = 0.6, p = 0.02) from hyperpolarized C-13 MRSI data.

CONCLUSIONS

Since the peak height in the ΔR2* curve from DSC data is related to the extent of blood volume, these hyperpolarized C-13 imaging parameters may be used to assess blood volume in rodent intracranial xenograft models of GBM.