Impact of SUSAN Denoising and ComBat Harmonization on Machine Learning Model Performance for Malignant Brain Neoplasms.

BACKGROUND AND PURPOSE: Feature variability in radiomics studies due to technical and magnet strength parameters is well known and may be addressed through various pre-processing methods. However, very few studies have evaluated downstream impact of variable pre-processing on model classification performance in a multi-class setting. We sought to evaluate the impact of SUSAN denoising and ComBat harmonization on model classification performance. MATERIALS AND METHODS: A total of 493 cases (410 internal and 83 external dataset) of glioblastoma (GB), intracranial metastatic disease (IMD) and primary CNS lymphoma (PCNSL) underwent semi-automated 3D-segmentation post baseline image processing (BIP) consisting of resampling, realignment, co-registration, skull stripping and image normalization. Post BIP, two sets were generated, one with and another without SUSAN denoising (SD). Radiomics features were extracted from both datasets and batch corrected to produce four datasets: (a) BIP, (b) BIP with SD, (c) BIP with ComBat and (d) BIP with both SD and ComBat harmonization. Performance was then summarized for models using a combination of six feature selection techniques and six machine learning models across four mask-sequence combinations with features derived from one-three (multi-parametric) MRI sequences. RESULTS: Most top performing models on the external test set used BIP+SD derived features. Overall, use of SD and ComBat harmonization led to a slight but generally consistent improvement in model performance on the external test set. CONCLUSIONS: The use of image pre-processing steps such as SD and ComBat harmonization may be more useful in a multiinstitutional setting and improve model generalizability. Models derived from only T1-CE images showed comparable performance to models derived from multiparametric MRI.

Imaging Findings Post-Stereotactic Radiosurgery for Vestibular Schwannoma: A Primer for the Radiologist.

Noninvasive tumor control of vestibular schwannomas through stereotactic radiosurgery allows high rates of long-term tumor control and has been used primarily for small- and medium-sized vestibular schwannomas. The posttreatment imaging appearance of the tumor, temporal patterns of growth and treatment response, as well as extratumoral complications can often be both subtle or confusing and should be appropriately recognized. Herein, the authors present an imaging-based review of expected changes as well as associated complications related to radiosurgery for vestibular schwannomas.

Molecular GBM versus Histopathological GBM: Radiology-Pathology-Genetic Correlation and the New WHO 2021 Definition of Glioblastoma.

Given the recent advances in molecular pathogenesis of tumors, with better correlation with tumor behavior and prognosis, major changes were made to the new 2021 WHO (CNS5) classification of CNS tumors, including updated criteria for diagnosis of glioblastoma. Diagnosis of GBM now requires absence of isocitrate dehydrogenase and histone 3 mutations (IDH-wildtype and H3-wildtype) as the basic cornerstone, with elimination of the IDH-mutated category. The requirements for diagnosis were conventionally histopathological, based on the presence of pathognomonic features such as microvascular proliferation and necrosis. However, even if these histological features are absent, many lower grade (WHO grade 2/3) diffuse astrocytic gliomas behave clinically similar to GBM (grade 4). The 2021 WHO classification introduced new molecular criteria that can be used to upgrade the diagnosis of such histologically lower-grade, IDH-wildtype, astrocytomas to GBM. The three molecular criteria include: concurrent gain of whole chromosome 7 and loss of whole chromosome 10 (+7/-10); TERT promoter mutation; epidermal growth factor receptor (EGFR) amplification. Given these changes, it is now strongly recommended to have molecular analysis of WHO grade 2/3 diffuse astrocytic, IDH-wildtype, gliomas in adult patients, as identification of any of the above mutations allows for upgrading the tumor to WHO grade 4 ("molecular GBM") with important prognostic implications. Despite at an early stage, there is active ongoing research on the unique MRI features of molecular GBM. This paper highlights the differences between "molecular" and "histopathological" GBM, with the aim of providing a basic understanding about these changes.ABBREVIATIONS: GBM=Glioblastoma; TERT=telomerase reverse transcriptase; EGFR=epidermal growth factor receptor; MGMT= methylguanine-DNA methyltransferase; NGS= next-generation sequencing; IDH= isocitrate dehydrogenase.

High-Grade Astrocytoma with Piloid Features: A Dual Institutional Review of Imaging Findings of a Novel Entity.

High-grade astrocytoma with piloid features (HGAP) is a recently identified brain tumor characterized by a distinct DNA methylation profile. Predominantly located in the posterior fossa of adults, HGAP is notably prevalent in individuals with neurofibromatosis type 1. We present an image-centric review of HGAP and explore the association between HGAP and neurofibromatosis type 1. Data were collected from 8 HGAP patients treated at two tertiary care institutions between January 2020 and October 2023. Demographic details, clinical records, management, and tumor molecular profiles were analyzed. Tumor characteristics, including location and imaging features on MR imaging, were reviewed. Clinical or imaging features suggestive of neurofibromatosis 1 or the presence of NF1 gene alteration were documented. The mean age at presentation was 45.5 years (male/female = 5:3). Tumors were midline, localized in the posterior fossa (n = 4), diencephalic/thalamic (n = 2), and spinal cord (n = 2). HGAP lesions were T1 hypointense, T2-hyperintense, mostly without diffusion restriction, predominantly peripheral irregular enhancement with central necrosis (n = 3) followed by mixed heterogeneous enhancement (n = 2). Two NF1 mutation carriers showed signs of neurofibromatosis type 1 before HGAP diagnosis, with one diagnosed during HGAP evaluation, strengthening the HGAP-NF1 link, particularly in patients with posterior fossa masses. All tumors were IDH1 wild-type, often with ATRX, CDKN2A/B, and NF1 gene alteration. Six patients underwent surgical resection followed by adjuvant chemoradiation. Six patients were alive, and two died during the last follow-up. Histone H3 mutations were not detected in our cohort, such as the common H3K27M typically seen in diffuse midline gliomas, linked to aggressive clinical behavior and poor prognosis. HGAP lesions may involve the brain or spine and tend to be midline or paramedian in location. Underlying neurofibromatosis type 1 diagnosis or imaging findings are important diagnostic cues.

Cauda equina neuroendocrine tumors: A single institutional imaging review of cases over two decades.

Cauda Equina Neuroendocrine Tumors (CE-NET), previously referred to as paragangliomas are a rare subset of spinal tumors, with limited data on imaging. Herein, we present a retrospective review of clinical and imaging findings of CE-NETs in ten patients who were evaluated at our institution over the past two decades. All patients had well-defined intradural lesions in the lumbar spine which demonstrated slow growth. A review of imaging findings revealed the presence of an eccentric vascular pedicle along the dorsal aspect of the tumor in 8 of the 10 patients (eccentric vessel sign), a distinctive finding that has not previously been reported with this tumor and may help improve the accuracy of imaging-based diagnosis. In all cases, a gross-total resection was performed, with resolution of symptoms in most of the cases.

Newly Recognized Genetic Tumor Syndromes of the CNS in the 5th WHO Classification: Imaging Overview with Genetic Updates.

The nervous system is commonly involved in a wide range of genetic tumor-predisposition syndromes. The classification of genetic tumor syndromes has evolved during the past years; however, it has now become clear that these syndromes can be categorized into a relatively small number of major mechanisms, which form the basis of the new 5th edition of the World Health Organization book (beta online version) on genetic tumor syndromes. For the first time, the World Health Organization has also included a separate chapter on genetic tumor syndromes in the latest edition of all the multisystem tumor series, including the 5th edition of CNS tumors. Our understanding of these syndromes has evolved rapidly since the previous edition (4th edition, 2016) with recognition of 8 new syndromes, including the following: Elongator protein complex-medulloblastoma syndrome, BRCA1-associated protein 1 tumor-predisposition syndrome, DICER1 syndrome, familial paraganglioma syndrome, melanoma-astrocytoma syndrome, Carney complex, Fanconi anemia, and familial retinoblastoma. This review provides a description of these new CNS tumor syndromes with a focus on imaging and genetic characteristics.

Clinical and Magnetic Resonance Imaging Radiomics-Based Survival Prediction in Glioblastoma Using Multiparametric Magnetic Resonance Imaging.

INTRODUCTION: Survival prediction in glioblastoma remains challenging, and identification of robust imaging markers could help with this relevant clinical problem. We evaluated multiparametric magnetic resonance imaging-derived radiomics to assess prediction of overall survival (OS) and progression-free survival (PFS). METHODOLOGY: A retrospective, institutional review board-approved study was performed. There were 93 eligible patients, of which 55 underwent gross tumor resection and chemoradiation (GTR-CR). Overall survival and PFS were assessed in the entire cohort and the GTR-CR cohort using multiple machine learning pipelines. A model based on multiple clinical variables was also developed. Survival prediction was assessed using the radiomics-only, clinical-only, and the radiomics and clinical combined models. RESULTS: For all patients combined, the clinical feature-derived model outperformed the best radiomics model for both OS (C-index, 0.706 vs 0.597; P < 0.0001) and PFS prediction (C-index, 0.675 vs 0.588; P < 0.001). Within the GTR-CR cohort, the radiomics model showed nonstatistically improved performance over the clinical model for predicting OS (C-index, 0.638 vs 0.588; P = 0.4). However, the radiomics model outperformed the clinical feature model for predicting PFS in GTR-CR cohort (C-index, 0.641 vs 0.550; P = 0.004). Combined clinical and radiomics model did not yield superior prediction when compared with the best model in each case. CONCLUSIONS: When considering all patients, regardless of therapy, the radiomics-derived prediction of OS and PFS is inferior to that from a model derived from clinical features alone. However, in patients with GTR-CR, radiomics-only model outperforms clinical feature-derived model for predicting PFS.

Differentiation Between Glioblastoma and Metastatic Disease on Conventional MRI Imaging Using 3D-Convolutional Neural Networks: Model Development and Validation.

RATIONALE AND OBJECTIVES: Imaging-based differentiation between glioblastoma (GB) and brain metastases (BM) remains challenging. Our aim was to evaluate the performance of 3D-convolutional neural networks (CNN) to address this binary classification problem. MATERIALS AND METHODS: T1-CE, T2WI, and FLAIR 3D-segmented masks of 307 patients (157 GB and 150 BM) were generated post resampling, co-registration normalization and semi-automated 3D-segmentation and used for internal model development. Subsequent external validation was performed on 59 cases (27 GB and 32 BM) from another institution. Four different mask-sequence combinations were evaluated using area under the curve (AUC), precision, recall and F1-scores. Diagnostic performance of a neuroradiologist and a general radiologist, both without and with the model output available, was also assessed. RESULTS: 3D-model using the T1-CE tumor mask (TM) showed the highest performance [AUC 0.93 (95% CI 0.858-0.995)] on the external test set, followed closely by the model using T1-CE TM and FLAIR mask of peri-tumoral region (PTR) [AUC of 0.91 (95% CI 0.834-0.986)]. Models using T2WI masks showed robust performance on the internal dataset but lower performance on the external set. Both neuroradiologist and general radiologist showed improved performance with model output provided [AUC increased from 0.89 to 0.968 (p = 0.06) and from 0.78 to 0.965 (p = 0.007) respectively], the latter being statistically significant. CONCLUSION: 3D-CNNs showed robust performance for differentiating GB from BMs, with T1-CE TM, either alone or combined with FLAIR-PTR masks. Availability of model output significantly improved the accuracy of the general radiologist.

AI-based classification of three common malignant tumors in neuro-oncology: A multi-institutional comparison of machine learning and deep learning methods.

PURPOSE: To determine if machine learning (ML) or deep learning (DL) pipelines perform better in AI-based three-class classification of glioblastoma (GBM), intracranial metastatic disease (IMD) and primary CNS lymphoma (PCNSL). METHODOLOGY: Retrospective analysis included 502 cases for training (208 GBM, 67 PCNSL and 227 IMD), with external validation on 86 cases (27:27:32). Multiparametric MRI images (T1W, T2W, FLAIR, DWI and T1-CE) were co-registered, resampled, denoised and intensity normalized, followed by semiautomatic 3D segmentation of the enhancing tumor (ET) and peritumoral region (PTR). Model performance was assessed using several ML pipelines and 3D-convolutional neural networks (3D-CNN) using sequence specific masks, as well as combination of masks. All pipelines were trained and evaluated with 5-fold nested cross-validation on internal data followed by external validation using multi-class AUC. RESULTS: Two ML models achieved similar performance on test set, one using T2-ET and T2-PTR masks (AUC: 0.885, 95% CI: [0.816, 0.935] and another using T1-CE-ET and FLAIR-PTR mask (AUC: 0.878, CI: [0.804, 0.930]). The best performing DL models achieved an AUC of 0.854, (CI [0.774, 0.914]) on external data using T1-CE-ET and T2-PTR masks, followed by model derived from T1-CE-ET, ADC-ET and FLAIR-PTR masks (AUC: 0.851, CI [0.772, 0.909]). CONCLUSION: Both ML and DL derived pipelines achieved similar performance. T1-CE mask was used in three of the top four overall models. Additionally, all four models had some mask derived from PTR, either T2WI or FLAIR.

Radiological abnormalities in progressive multifocal leukoencephalopathy: Identifying typical and atypical imaging patterns for early diagnosis and differential considerations.

Progressive multifocal leukoencephalopathy (PML) is a rare viral central nervous system (CNS) demyelinating disease primarily associated with a compromised immune system. PML is seen mainly in individuals with human immunodeficiency virus, lymphoproliferative disease, and multiple sclerosis. Patients on immunomodulators, chemotherapy, and solid organ or bone marrow transplants are predisposed to PML. Recognition of various PML-associated typical and atypical imaging abnormalities is critical for early diagnosis and differentiating it from other conditions, especially in high-risk populations. Early PML recognition should expedite efforts at immune-system restoration, allowing for a favorable outcome. This review aims to provide a practical overview of radiological abnormalities in PML patients and address differential considerations.

Fibroblast Activation Protein Inhibitor-Based Radionuclide Therapies: Current Status and Future Directions.

Metastatic malignancies have limited management strategies and variable treatment responses. Cancer cells develop beside and depend on the complex tumor microenvironment. Cancer-associated fibroblasts, with their complex interaction with tumor and immune cells, are involved in various steps of tumorigenesis, such as growth, invasion, metastasis, and treatment resistance. Prooncogenic cancer-associated fibroblasts emerged as attractive therapeutic targets. However, clinical trials have achieved suboptimal success. Fibroblast activation protein (FAP) inhibitor-based molecular imaging has shown encouraging results in cancer diagnosis, making them innovative targets for FAP inhibitor-based radionuclide therapies. This review summarizes the results of preclinical and clinical FAP-based radionuclide therapies. We will describe advances and FAP molecule modification in this novel therapy, as well as its dosimetry, safety profile, and efficacy. This summary may guide future research directions and optimize clinical decision-making in this emerging field.

Evaluating Autoencoders for Dimensionality Reduction of MRI-derived Radiomics and Classification of Malignant Brain Tumors.

Malignant brain tumors including parenchymal metastatic (MET) lesions, glioblastomas (GBM), and lymphomas (LYM) account for 29.7% of brain cancers. However, the characterization of these tumors from MRI imaging is difficult due to the similarity of their radiologically observed image features. Radiomics is the extraction of quantitative imaging features to characterize tumor intensity, shape, and texture. Applying machine learning over radiomic features could aid diagnostics by improving the classification of these common brain tumors. However, since the number of radiomic features is typically larger than the number of patients in the study, dimensionality reduction is needed to balance feature dimensionality and model complexity. Autoencoders are a form of unsupervised representation learning that can be used for dimensionality reduction. It is similar to PCA but uses a more complex and non-linear model to learn a compact latent space. In this work, we examine the effectiveness of autoencoders for dimensionality reduction on the radiomic feature space of multiparametric MRI images and the classification of malignant brain tumors: GBM, LYM, and MET. We further aim to address the class imbalances imposed by the rarity of lymphomas by examining different approaches to increase overall predictive performance through multiclass decomposition strategies.

CT Perfusion Maps Improve Detection of M2-MCA Occlusions in Acute Ischemic Stroke.

OBJECTIVES: Middle cerebral artery occlusions, particularly M2 branch occlusions are challenging to identify on CTA. We hypothesized that additional review of the CTP maps will increase large vessel occlusion (LVO) detection accuracy on CTA and reduce interpretation time. MATERIALS AND METHODS: Two readers (R1 and R2) retrospectively reviewed the CT studies in 99 patients (27 normal, 26 M1-MCA, 46 M2-MCA occlusions) who presented with suspected acute ischemic stroke (AIS). The time of interpretation and final diagnosis were recorded for the CTA images (derived from CTP data), both without and with the CTP maps. The time for analysis for all vascular occlusions was compared using McNemar tests. ROC curve analysis and McNemar tests were performed to assess changes in diagnostic performance with the addition of CTP maps. RESULTS: With the addition of the CTP maps, both readers showed increased sensitivity (p = 0.01 for R1 and p = 0.04 for R2), and accuracy (p = 0.02 for R1 and p = 0.004 for R2) for M2-MCA occlusions. There was a significant improvement in diagnostic performance for both readers for detection of M2-MCA occlusions (AUC R1 = 0.86 to 0.95, R2 = 0.84 to 0.95; p < 0.05). Both readers showed reduced interpretation time for all cases combined, as well as for normal studies (p < 0.001) when CTP images were reviewed along with CTA. Both readers also showed reduced interpretation time for M2-MCA occlusions, which was significant for one of the readers (p < 0.02). CONCLUSION: The addition of CTP maps improves accuracy and reduces interpretation time for detecting LVO and M2-MCA occlusions in AIS. Incorporation of CTP in acute stroke imaging protocols may improve detection of more distal occlusions.

Long-term performance of untreated fresh autologous pericardium as a valve substitute in pulmonary position.

Background: Pulmonary regurgitation is imminent after transannular patch (TAP). We analyze the long-term performance of untreated autologous pericardium (UAP) as valve substitute at pulmonary position in patients requiring TAP. Material and Methods: This cross-sectional study include patients operated between 2007 and 2012 (n = 92). A sample of 19 patients was selected for this study which had a follow-up of more than 3 years. This includes patients with no TAP (n = 4) and with TAP and valve substitute, a monocusp (n = 11) or a tricuspid valve (n = 4) at neopulmonary annulus. Patients underwent echocardiography for assessment of right ventricle function and 18 fluoro-deoxyglucose PET CT scan for measurements of valve substitute at neopulmonary annulus. The target to blood ratio (TBR) of uptake of glucose by monocusp was measured at the cooptation edge of the neopulmonary valve. Results: The median age of the patients is 14 (9 - 37). RV function is preserved (TAPSE 18.9 (10.6 - 22.8)) at a mean follow-up of 4 years (3-9). The measurements of monocusp shows a shrinkage in height of the cusp by 35.5% (70% - 1.0%) and length by 7% (-44% - +104%). There was less shrinkage observed in patients below 15 years of age. The TBR of monocusp was 0.945 (0.17 - 3.35) with a strong correlation between the TBR values of aortic valve leaflet and monocusp leaflet of same patient. Conclusion: The UAP is functional and successful as a valve substitute at neo pulmonary annulus at long-term follow-up. It has resisted calcification and has shown uptake of glucose in physiological limits.

Multimodality imaging in a case of multiple pulmonary hyalinizing granulomas - A decade follow-up.

A 44-year-old male was referred to our clinic (2015) to evaluate multiple lung nodules with increasing fatigue, dyspnea, and weight loss. He was being assessed to an outside hospital for the same since 2010. The X-ray and computed-tomography (CT)-chest showed numerous pulmonary nodules and bilateral hilar adenopathy. Imaging workup at our institute (2015) redemonstrated extensive calcified pulmonary nodules. 18fluoro-2-deoxy-d-glucose positron emission tomographyCT showed widespread pulmonary nodules with low-grade uptake. Video-assisted thoracic surgery lung biopsy revealed pulmonary hyalinizing granuloma (PHG). Recently because of increasing symptoms, he is being evaluated for a lung transplant. This case represents a rare diagnosis of PHG with a decade follow-up.

Vascular Involvement in Neurosarcoidosis: Early Experiences From Intracranial Vessel Wall Imaging.

BACKGROUND AND OBJECTIVES: Cerebrovascular manifestations in neurosarcoidosis (NS) were previously considered rare but are being increasingly recognized. We report our preliminary experience in patients with NS who underwent high-resolution vessel wall imaging (VWI). METHODS: A total of 13 consecutive patients with NS underwent VWI. Images were analyzed by 2 neuroradiologists in consensus. The assessment included segment-wise evaluation of larger- and medium-sized vessels (internal carotid artery, M1-M3 middle cerebral artery; A1-A3 anterior cerebral artery; V4 segments of vertebral arteries; basilar artery; and P1-P3 posterior cerebral artery), lenticulostriate perforator vessels, and medullary and deep cerebral veins. Cortical veins were not assessed due to flow-related artifacts. Brain biopsy findings were available in 6 cases and were also reviewed. RESULTS: Mean patient age was 54.9 years (33-71 years) with an M:F of 8:5. Mean duration between initial diagnosis and VWI study was 18 months. Overall, 9/13 (69%) patients had vascular abnormalities. Circumferential large vessel enhancement was seen in 3/13 (23%) patients, whereas perforator vessel involvement was seen in 6/13 (46%) patients. Medullary and deep vein involvement was also seen in 6/13 patients. In addition, 7/13 (54%) patients had microhemorrhages in susceptibility-weighted imaging, and 4/13 (31%) had chronic infarcts. On biopsy, 5/6 cases showed perivascular granulomas with vessel wall involvement in all 5 cases. DISCUSSION: Our preliminary findings suggest that involvement of intracranial vascular structures may be a common finding in patients with NS and should be routinely looked for. These findings appear concordant with previously reported autopsy literature and need to be validated on a larger scale.

Radiomic Based Machine Learning Performance for a Three Class Problem in Neuro-Oncology: Time to Test the Waters?

Prior radiomics studies have focused on two-class brain tumor classification, which limits generalizability. The performance of radiomics in differentiating the three most common malignant brain tumors (glioblastoma (GBM), primary central nervous system lymphoma (PCNSL), and metastatic disease) is assessed; factors affecting the model performance and usefulness of a single sequence versus multiparametric MRI (MP-MRI) remain largely unaddressed. This retrospective study included 253 patients (120 metastatic (lung and brain), 40 PCNSL, and 93 GBM). Radiomic features were extracted for whole a tumor mask (enhancing plus necrotic) and an edema mask (first pipeline), as well as for separate enhancing and necrotic and edema masks (second pipeline). Model performance was evaluated using MP-MRI, individual sequences, and the T1 contrast enhanced (T1-CE) sequence without the edema mask across 45 model/feature selection combinations. The second pipeline showed significantly high performance across all combinations (Brier score: 0.311-0.325). GBRM fit using the full feature set from the T1-CE sequence was the best model. The majority of the top models were built using a full feature set and inbuilt feature selection. No significant difference was seen between the top-performing models for MP-MRI (AUC 0.910) and T1-CE sequence with (AUC 0.908) and without edema masks (AUC 0.894). T1-CE is the single best sequence with comparable performance to that of multiparametric MRI (MP-MRI). Model performance varies based on tumor subregion and the combination of model/feature selection methods.

Role of FDG PET/CT for detection of primary tumor in patients with extracervical metastases from carcinoma of unknown primary.

AIM: To explore the diagnostic performance of 18F-fluorodeoxyglucose positron emission tomography/computed tomography (FDG PET/CT) to detect the primary tumor site in patients with extracervical metastases from carcinoma of unknown primary (CUP). We evaluated patient outcomes as overall survival (OS). MATERIALS AND METHODS: In a single-center, retrospective study (2005-2019), patients with extracervical metastases from CUP underwent FDG PET/CT to detect primary tumor sites. The final diagnosis was based on histopathology/or clinical follow-up of at least 12 months. RESULTS: A total of 83 patients [Male 41 (49%), mean age 59 ± 14 years, range: 32-83 years] fulfilled the inclusion/exclusion criteria and were enrolled for analysis. The primary tumor was detected in 36 out of 83 (43%) patients based on histopathology/or clinical follow-up. PET/CT suggested the primary tumor site in 39 (47%) patients with diagnostic accuracy of 87%, sensitivity 89%, specificity 85%, PPV 82%, NPV 91% and detection rate 39%. Patients with oligometastases (<3) (2.16 years, 1.04-2.54) and primary unidentified (1 year, 0.34-2.14) had longer median survival time compared to the patients with multiple metastases (0.67 years, 0.17-1.58, p = 0.009) and primary identified (0.67 years,0.16-1.33, p = 0.002). The SUVmax of the primary or metastatic lesions with maximum uptake was not significantly related to survival. CONCLUSIONS: PET/CT could reveal the primary tumor site in 39% of the patients. It demonstrated the metastatic disease burden and distribution in patients with 'primary obscured', which directs management. Patients with multiple metastases and primary identified had a poorer prognosis. In patients with primary unidentified after PET/CT, a further search was futile.

Machine learning based differentiation of glioblastoma from brain metastasis using MRI derived radiomics.

Few studies have addressed radiomics based differentiation of Glioblastoma (GBM) and intracranial metastatic disease (IMD). However, the effect of different tumor masks, comparison of single versus multiparametric MRI (mp-MRI) or select combination of sequences remains undefined. We cross-compared multiple radiomics based machine learning (ML) models using mp-MRI to determine optimized configurations. Our retrospective study included 60 GBM and 60 IMD patients. Forty-five combinations of ML models and feature reduction strategies were assessed for features extracted from whole tumor and edema masks using mp-MRI [T1W, T2W, T1-contrast enhanced (T1-CE), ADC, FLAIR], individual MRI sequences and combined T1-CE and FLAIR sequences. Model performance was assessed using receiver operating characteristic curve. For mp-MRI, the best model was LASSO model fit using full feature set (AUC 0.953). FLAIR was the best individual sequence (LASSO-full feature set, AUC 0.951). For combined T1-CE/FLAIR sequence, adaBoost-full feature set was the best performer (AUC 0.951). No significant difference was seen between top models across all scenarios, including models using FLAIR only, mp-MRI and combined T1-CE/FLAIR sequence. Top features were extracted from both the whole tumor and edema masks. Shape sphericity is an important discriminating feature.