Faculty Bibliography

Machine learning (ML) models have been shown to predict the presence of clinical factors from medical imaging with remarkable accuracy. However, these complex models can be difficult to interpret and are often criticized as "black boxes". Prediction models that provide no insight into how their predictions are obtained are difficult to trust for making important clinical decisions, such as medical diagnoses or treatment. Explainable machine learning (XML) methods, such as Shapley values, have made it possible to explain the behavior of ML algorithms and to identify which predictors contribute most to a prediction. Incorporating XML methods into medical software tools has the potential to increase trust in ML-powered predictions and aid physicians in making medical decisions. Specifically, in the field of medical imaging analysis the most used methods for explaining deep learning-based model predictions are saliency maps that highlight important areas of an image. However, they do not provide a straightforward interpretation of which qualities of an image area are important. Here, we describe a novel pipeline for XML imaging that uses radiomics data and Shapley values as tools to explain outcome predictions from complex prediction models built with medical imaging with well-defined predictors. We present a visualization of XML imaging results in a clinician-focused dashboard that can be generalized to various settings. We demonstrate the use of this workflow for developing and explaining a prediction model using MRI data from glioma patients to predict a genetic mutation.

BACKGROUND AND PURPOSE/OBJECTIVE:-mutant 1p/19q noncodeleted gliomas with a high positive predictive value. We have developed an approach to quantify the T2-FLAIR mismatch signature and use it to predict the molecular status of lower-grade gliomas. MATERIALS AND METHODS/METHODS:We used multiparametric MR imaging scans and segmentation labels of 108 preoperative lower-grade glioma tumors from The Cancer Imaging Archive. Clinical information and T2-FLAIR mismatch sign labels were obtained from supplementary material of relevant publications. We adopted an objective analytic approach to estimate this sign through a geographically weighted regression and used the residuals for each case to construct a probability density function (serving as a residual signature). These functions were then analyzed using an appropriate statistical framework. RESULTS:-mutant 1p/19q noncodeleted class of tumors versus other categories. Our classifier predicts these cases with area under the curve of 0.98 and high specificity and sensitivity. It also predicts the T2-FLAIR mismatch sign within these cases with an under the curve of 0.93. CONCLUSIONS:-mutation and 1p/19q codeletion status with high predictive power. The utility of the proposed quantification of the T2-FLAIR mismatch sign can be potentially validated through a prospective multi-institutional study.

BACKGROUND AND PURPOSE/OBJECTIVE:We reviewed the literature to evaluate cerebrospinal fluid (CSF) results from patients with coronavirus disease 2019 (COVID-19) who had neurological symptoms and had an MRI that showed (1) central nervous system (CNS) hyperintense lesions not attributed to ischemia and/or (2) leptomeningeal enhancement. We sought to determine if these findings were associated with a positive CSF severe acute respiratory syndrome associated coronavirus 2 (SARS-CoV-2) polymerase chain reaction (PCR). METHODS:We performed a systematic review of Medline and Embase from December 1, 2019 to November 18, 2020. CSF results were evaluated based on the presence/absence of (1) ≥ 1 CNS hyperintense lesion and (2) leptomeningeal enhancement. RESULTS:In 117 publications, we identified 193 patients with COVID-19 who had an MRI of the CNS and CSF testing. There were 125 (65%) patients with CNS hyperintense lesions. Patients with CNS hyperintense lesions were significantly more likely to have a positive CSF SARS-CoV-2 PCR (10% [9/87] vs. 0% [0/43], p = 0.029). Of 75 patients who had a contrast MRI, there were 20 (27%) patients who had leptomeningeal enhancement. Patients with leptomeningeal enhancement were significantly more likely to have a positive CSF SARS-CoV-2 PCR (25% [4/16] vs. 5% [2/42], p = 0.024). CONCLUSION/CONCLUSIONS:The presence of CNS hyperintense lesions or leptomeningeal enhancement on neuroimaging from patients with COVID-19 is associated with increased likelihood of a positive CSF SARS-CoV-2 PCR. However, a positive CSF SARS-CoV-2 PCR is uncommon in patients with these neuroimaging findings, suggesting they are often related to other etiologies, such as inflammation, hypoxia, or ischemia.

BACKGROUND:Evolution of brain magnetic resonance imaging (MRI) findings in critically ill patients with coronavirus disease 2019 (COVID-19) is unknown. METHODS:We retrospectively reviewed 4530 critically ill patients with COVID-19 admitted to three tertiary care hospitals in New York City from March 1 to June 30, 2020 to identify patients who had more than one brain MRI. We reviewed the initial and final MRI for each patient to (1) measure the percent change in the bicaudate index and third ventricular diameter and (2) evaluate changes in the presence and severity of white matter changes. RESULTS:Twenty-one patients had two MRIs separated by a median of 22 [Interquartile range (IQR) 14-30] days. Ventricle size increased for 15 patients (71%) between scans [median bicaudate index 0.16 (IQR 0.126-0.181) initially and 0.167 (IQR 0.138-0.203) on final imaging (p < 0.001); median third ventricular diameter 6.9 mm (IQR 5.4-10.3) initially and 7.2 mm (IQR 6.4-10.8) on final imaging (p < 0.001)]. Every patient had white matter changes on the initial and final MRI; between images, they worsened for seven patients (33%) and improved for three (14%). CONCLUSIONS:On serial imaging of critically ill patients with COVID-19, ventricle size frequently increased over several weeks. White matter changes were often unchanged, but in some cases they worsened or improved, demonstrating there is likely a spectrum of pathophysiological processes responsible for these changes.

PURPOSE/OBJECTIVE:The WHO 2016 update classifies glioblastomas (WHO grade IV) according to isocitrate dehydrogenase (IDH) gene mutation status. We aimed to determine MRI-based metrics for predicting IDH mutation in glioblastoma. METHODS:This retrospective study included glioblastoma cases (n = 199) with known IDH mutation status and pre-operative MRI (T1WI, T2WI, FLAIR, contrast-enhanced T1W1 at minimum). Two neuroradiologists determined the following MRI metrics: (1) primary lobe of involvement (frontal or non-frontal); (2) presence/absence of contrast-enhancement; (3) presence/absence of necrosis; (4) presence/absence of fluid attenuation in the non-contrast-enhancing tumor (nCET); (5) maximum width of peritumoral edema (cm); (6) presence/absence of multifocal disease. Inter-reader agreement was determined. After resolving discordant measurements, multivariate association between consensus MRI metrics/patient age and IDH mutation status was determined. RESULTS:Among 199 glioblastomas, 16 were IDH-mutant. Inter-reader agreement was calculated for contrast-enhancement (ĸ = 0.49 [- 0.11-1.00]), necrosis (ĸ = 0.55 [0.34-0.76]), fluid attenuation in nCET (ĸ = 0.83 [0.68-0.99]), multifocal disease (ĸ = 0.55 [0.39-0.70]), and primary lobe (ĸ = 0.85 [0.80-0.91]). Mean difference for peritumoral edema width between readers was 0.3 cm [0.2-0.5], p < 0.001. Multivariate analysis uncovered significant associations between IDH-mutation and fluid attenuation in nCET (OR 82.9 [19.22, ∞], p < 0.001), younger age (OR 0.93 [0.86, 0.98], p = 0.009), frontal lobe location (OR 11.08 [1.14, 352.97], p = 0.037), and less peritumoral edema (OR 0.15 [0, 0.65], p = 0.044). CONCLUSIONS:Conventional MRI metrics and patient age predict IDH-mutation status in glioblastoma. Among MRI markers, fluid attenuation in nCET represents a novel marker with high inter-reader agreement that is strongly associated with Glioblastoma, IDH-mutant.

PURPOSE/OBJECTIVE:Resting state functional magnetic resonance imaging (rsfMRI) is an emerging tool to explore the functional connectivity of different brain regions. We aimed to assess the disruption of functional connectivity of the Default Mode Network (DMN), Dorsal Attention Network(DAN) and Fronto-Parietal Network (FPN) in patients with glial tumors. METHODS:rsfMRI data acquired on 3T-MR of treatment-naive glioma patients prospectively recruited (2015-2019) and matched controls from the 1000 functional-connectomes-project were analyzed using the CONN functional toolbox. Seed-Based Connectivity Analysis (SBCA) and Independent Component Analysis (ICA, with 10 to 100 components) were performed to study reliably the three networks of interest. RESULTS:). For the FPN, increased connectivity was noted in the precuneus, posterior cingulate gyrus, and frontal cortex. No difference in the connectivity of the networks of interest was demonstrated between low- and high-grade gliomas, as well as when stratified by their IDH1-R132H (isocitrate dehydrogenase) mutation status. CONCLUSION/CONCLUSIONS:Altered functional connectivity is reliably found with SBCA and ICA in the DMN, DAN, and FPN in glioma patients, possibly explained by decreased connectivity between the cerebral hemispheres across the corpus callosum due to disruption of the connections.