TABLE 3.
Summary of selected studies on imaging approaches for identifying brain metastases
| Technique | Main findings | References |
|---|---|---|
| Single‐voxel proton MR spectroscopy | The presence of intratumoral creatine indicates GBM, whereas its absence indicates metastasis, in samples consisting of 11 anaplastic gliomas, 20 GBMs and 25 metastases | [67] |
| MRI | No differentiation between GBM tumors and brain metastases when using the mean apparent diffusion coefficient and absolute standard deviation derived from apparent diffusion coefficient measurements based on cellularity levels (n = 34 patients) | [68] |
| MRI | Use of the apparent diffusion coefficient could differentiate GBM from metastasis | [69] |
| Higher homogeneity and inverse difference moment in GBM compared to metastases (n = 36 GBMs and 26 metastases) | ||
| MRI | Heterogeneity of the GBMs peritumoral edema was significantly higher than the edema surrounding MET by texture analysis, allowing a differentiation with a sensitivity of 80% and specificity of 90% (n = 22 GBM tumors and 21 metastases) | [70] |
| MRI | Combining arterial spin labeling perfusion (ASL)‐ and diffusion tensor imaging (DTI)‐derived metrics could differentiate GBM from brain metastases (n = 36 patients with provisional diagnosis of GBM or metastasis) | [71] |
| MRI | The use of 2D texture features extracted from images obtained with MRI enable the discrimination between GBM and brain metastases (n = 50 patients with GBM and 50 with metastasis) | [72] |
| MRI | Computational‐aided quantitative analysis of MRI images showed high accuracy in differentiating GBM from metastases, with texture features being more relevant than fractal‐based features (n = 30 patients with GBM and 25 with metastasis) | [73] |
| MRI | High sensitivity and specificity obtained in the distinction between GBM and solitary brain metastases with the use of post‐contrast 3 DT1 MR images selection and optimized by a machine learning classifier with a nested cross‐validation (n = 71 patients with GBM and 72 with metastasis) | [74] |
| MRI | Development of an efficient deep learning‐based model validated using images from 498 patients, from a total sample of records from 598 patients with confirmed GBM or metastasis analyzed retrospectively | [75] |
| MRI | A trained multi‐class multi‐layer perceptron model using parameters from preoperative MR images could differentiate GBM, brain metastasis and central nervous system lymphoma provides approximately 19% increase in diagnostic yield | [76] |
| MRI | Multi‐class multi‐player models trained with tumor volumes could discriminate among GBM, brain metastasis and central nervous system lymphoma with a maximum accuracy of 69.2%. Using the MLP model as a computer‐aided diagnosis for cases in which human reviewers disagreed on the diagnosis resulted in correct diagnoses 19.2% additional cases (n = GBM, 9 metastasis; and 8 central nervous system lymphoma | [77] |
| High‐resolution confocal laser endoscopic images | A trained residual network model that allows automated, on‐site analysis of tumor specimens based on confocal laser endoscopic image data sets achieved a test accuracy of 90.9% after applying a two‐class prediction for GBMs versus brain metastases, whereas for three class predictions the model achieved a ratio of correct predicted label of 85.8% in the test set | [78] |
| A prediction accuracy to 98.6% was reached by applying a confidence rate of 0.999 when images with substantial artifacts were excluded (n = 25 patients with GBM, brain metastasis or meningioma) | ||
| Contrast perfusion MRI | The cerebral blood volume gradient in the peritumoral brain zone was the most efficient parameter to differentiate GBM tumors from metastases | [79] |
| Phase difference enhanced imaging (PADRE) MRI | Evaluation of peritumoral areas on Color PADRE helped to differentiate GBM tumors from metastases (n = 17 patients with GBM, 11 patients with brain metastases and 9 with diffuse astrocytoma) | [80] |
| PET with use of sequential tracers | The use of [11C]methionine as a PET tracer could identify GBM regrowth in all lesions examined, whereas [11C]PBR28 could only identify 3 of 7 lesions (n = 5 patients previously treated brain metastases showing regrowth) | [81] |
Abbreviations: MR = magnetic resonance; DT1 = diffusion tensor imaging (DT1); GBM = glioblastoma; MRI = magnetic resonance imaging; PADRE = map of phase difference enhanced imaging.