Education & Certifications
Bachelor of Science, Massachusetts Institute of Technology, Brain and Cognitive Sciences (2017)
Machine Learning Approach to Differentiation of Peripheral Schwannomas and Neurofibromas: A Multi-Center Study.
BACKGROUND: Non-invasive differentiation between schwannomas and neurofibromas is important for appropriate management, preoperative counseling, and surgical planning, but has proven difficult using conventional imaging. The objective of this study was to develop and evaluate machine learning approaches for differentiating peripheral schwannomas from neurofibromas.METHODS: We assembled a cohort of schwannomas and neurofibromas from 3 independent institutions and extracted high-dimensional radiomic features from gadolinium-enhanced, T1-weighted MRI using the PyRadiomics package on Quantitative Imaging Feature Pipeline. Age, sex, neurogenetic syndrome, spontaneous pain, and motor deficit were recorded. We evaluated the performance of 6 radiomics-based classifier models with and without clinical features and compared model performance against human expert evaluators.RESULTS: 107 schwannomas and 59 neurofibroma were included. The primary models included both clinical and imaging data. The accuracy of the human evaluators (0.765) did not significantly exceed the no-information rate (NIR), whereas the Support Vector Machine (0.929), Logistic Regression (0.929), and Random Forest (0.905) classifiers exceeded the NIR. Using the method of DeLong, the AUC for the Logistic Regression (AUC=0.923) and K Nearest Neighbor (AUC=0.923) classifiers was significantly greater than the human evaluators (AUC=0.766; p = 0.041).CONCLUSIONS: The radiomics-based classifiers developed here proved to be more accurate and had a higher AUC on the ROC curve than expert human evaluators. This demonstrates that radiomics using routine MRI sequences and clinical features can aid in differentiation of peripheral schwannomas and neurofibromas.
View details for DOI 10.1093/neuonc/noab211
View details for PubMedID 34487172
Acetazolamide-Challenged Arterial Spin Labeling Detects Augmented Cerebrovascular Reserve After Surgery for Moyamoya
View details for DOI 10.1161/STROKEAHA.121.036616
Validation of the Global Limb Anatomical Staging System in First-time Lower Extremity Revascularization.
Journal of vascular surgery
The Global Limb Anatomical Staging System (GLASS) was developed as a new anatomic classification scheme for grading the severity of chronic limb threatening ischemia (CLTI). We evaluated the ability of this anatomic grading system to determine major adverse limb events following lower extremity revascularization.We performed a single-institutional retrospective review of 1,060 consecutive patients undergoing 1,180 first-time open or endovascular revascularization procedures for CLTI from 2005-2014. Based on review of angiographic images, limbs were classified as GLASS Stage 1, 2, or 3. The primary composite outcome was reintervention, major amputation (below or above knee amputation), or restenosis (>3.5x step-up by duplex criteria) events (RAS). Secondary outcomes included all-cause mortality, failure to cross the lesion by endovascular methods, and comparison between bypass versus endovascular intervention. Kaplan-Meier estimates were used to determine event rates at 1- and 5-years and Cox regression analysis to adjust for baseline differences among the GLASS stages.Of all patients undergoing first-time revascularization, 1,180 patients (91%) had imaging available for GLASS grading, of which 552 limbs were treated with open bypass (47%) and 628 limbs with endovascular intervention (53%). Compared with GLASS Stage 1 disease (N=267, 23%), Stage 2 (N=367, 31%) and Stage 3 (N=546, 42%) were associated with higher risk of RAS at 1-year (Stage 1: 33% vs. Stage 2: 48% vs. Stage 3: 53%) and 5-year follow-up (Stage 1: 45%, reference; Stage 2: 65%, HR 1.7 [1.3-2.2], P < .001; Stage 3: 69%, HR 2.3 [1.7-2.9], P < .001). These differences were mainly driven by reintervention and restenosis rather than by major amputation. Five-year mortality was similar in Stage 2 and 3 compared with Stage 1 disease (Stage 1: 40%, reference; Stage 2: 45%, HR 1.1 [0.8-1.4], P = .69; Stage 3: 49%, HR 1.2 [1.0-1.6], P = .11). For all attempted endovascular interventions, failure to cross a target lesion increased with advancing GLASS Stage (Stage 1: 4.5% vs. Stage 2: 6.3% vs. Stage 3: 13.3%, P < .01). Compared with open bypass (N=552, 46.8%), endovascular intervention (N=628, P=53.3%) was associated with a higher rate of 5-year RAS for GLASS Stage 1 (49% vs 34%; HR 1.9 [1.1-3.5], P=.03), Stage 2 (69% vs 52%, HR 1.7 [1.2-2.5], P < .01), and Stage 3 disease (83% vs 61%, HR 1.5 [1.2-2.0], P < .01).In patients undergoing first-time lower extremity revascularization, the GLASS anatomic classification scheme can be used to predict reintervention and restenosis. Bypass has better long-term outcomes compared with endovascular intervention for all GLASS stages.
View details for DOI 10.1016/j.jvs.2020.08.151
View details for PubMedID 33091516
Collateral status contributes to differences between observed and predicted 24-h infarct volumes in DEFUSE 3.
Journal of cerebral blood flow and metabolism : official journal of the International Society of Cerebral Blood Flow and Metabolism
We previously demonstrated that in the DEFUSE 3 trial, the union of the baseline core and the 24-h Tmax>6s perfusion lesion predicts the infarct volume at 24h. Presently, we assessed if collateral robustness measured by the hypoperfusion intensity ratio (HIR) and cerebral blood volume (CBV) index accounts for the variance in these predictions. DEFUSE 3 patients underwent MRI/CT perfusion imaging at baseline and 24h post-randomization. We compared baseline and follow-up HIR and CBV index across subgroups stratified by differences between predicted and observed 24-h infarct volumes. Of 123 eligible patients, 34 with 24-h infarcts larger than predicted had less favorable collaterals at baseline (HIR 0.43 vs. 0.32, p=0.006; CBV Index 0.78 vs. 0.85, p=0.001) and 24h (HIR 0.56 vs. 0.07, p=0.004; CBV Index 0.47 vs. 0.73, p=0.006) compared to 71 patients with more accurate infarct volume prediction. Eighteen patients with 24-h infarcts smaller than predicted had similar baseline collateral scores but more favorable 24-h CBV indices (0.81 vs. 0.73, p=0.040). Overall, patients with 24-h infarcts larger than predicted had evidence of less favorable baseline collaterals that fail within 24h, while patients with 24-h infarcts smaller than predicted typically had favorable collaterals that persisted for 24h.
View details for DOI 10.1177/0271678X20918816
View details for PubMedID 32423329
- Ischemic Core and Hypoperfusion Volumes Correlate With Infarct Size 24 Hours After Randomization in DEFUSE 3 STROKE 2019; 50 (3): 626–31
Contemporary outcomes after carotid endarterectomy in high-risk anatomic and physiologic patients.
Journal of vascular surgery
Current guidelines state that the acceptable 30-day postoperative stroke/death rate after carotid endarterectomy (CEA) is <3% for asymptomatic patients and <6% for symptomatic patients. The Centers for Medicare and Medicaid Services has identified certain high-risk characteristics used to define patients at highest risk for CEA for whom carotid artery stenting would be reimbursed. We evaluated the impact of the Centers for Medicare and Medicaid Services physiologic and anatomic high-risk criteria on major adverse event rates after CEA in asymptomatic and symptomatic patients.We retrospectively reviewed all patients undergoing CEA from 2011 to 2017 in the American College of Surgeons National Surgical Quality Improvement Program vascular targeted database. Patients with high-risk anatomic or physiologic characteristics were identified by a predefined variable and were compared with normal-risk patients. The primary outcome was 30-day stroke/death, stratified by symptom status.We identified 25,788 patients undergoing CEA, of whom 60% were treated for asymptomatic carotid disease. Among all patients, high-risk physiology or anatomy was associated with higher rates of 30-day stroke/death compared with normal-risk patients (physiologic risk, 4.6% vs 2.3% [P < .001]; anatomic risk, 3.6% vs 2.3% [P < .001]). Patients who met criteria for high-risk physiology or anatomy also had higher rates of cardiac events (physiologic risk, 3.1% vs 1.6% [P < .001]; anatomic risk, 2.3% vs 1.6% [P < .01]), but only patients with high-risk anatomy had higher rates of cranial nerve injury (physiologic risk, 2.4% vs 2.5% [P = .81]; anatomic risk, 4.3% vs 2.5% [P < .001]). Asymptomatic patients with high-risk physiology or anatomy had higher rates of 30-day stroke/death, especially in the physiologic high-risk group (physiologic risk, 4.7% vs 1.5% [P < .001]; anatomic risk, 2.6% vs 1.5% [P < .01]), compared with normal-risk patients. However, among symptomatic patients, differences in stroke/death were seen only with high-risk anatomic patients and not with high-risk physiologic patients (physiologic risk, 4.6% vs 3.4% [P = .12]; anatomic risk, 4.8% vs 3.4% [P = .01]).As currently selected, contemporary real-world outcomes after CEA in asymptomatic carotid disease patients meeting high-risk physiologic criteria show an unacceptably high 30-day stroke/death rate, well above the 3% threshold. These results suggest the need for better selection of patients and preoperative optimization before elective CEA.
View details for DOI 10.1016/j.jvs.2019.05.041
View details for PubMedID 31443978
Vascular Risk and β-Amyloid Are Synergistically Associated with Cortical Tau.
Annals of neurology
2019; 85 (2): 272–79
Neuropathological studies have demonstrated that cerebrovascular disease and Alzheimer disease (AD) pathology frequently co-occur in older adults. The extent to which cerebrovascular disease influences the progression of AD pathology remains unclear. Leveraging newly available positron emission tomography (PET) imaging, we examined whether a well-validated measure of systemic vascular risk and β-amyloid (Aβ) burden have an interactive association with regional tau burden.Vascular risk was quantified at baseline in 152 clinically normal older adults (mean age = 73.5 ± 6.1 years) with the office-based Framingham Heart Study cardiovascular disease risk algorithm (FHS-CVD). We acquired Aβ (11 C-Pittsburgh compound B) and tau (18 F-flortaucipir) PET imaging on the same participants. Aβ PET was performed at baseline; tau PET was acquired on average 2.98 ± 1.1 years later. Tau was measured in the entorhinal cortex (EC), an early site of tau deposition, and in the inferior temporal cortex (ITC), an early site of neocortical tau accumulation associated with AD. Linear regression models examined FHS-CVD and Aβ as interactive predictors of tau deposition, adjusting for age, sex, APOE ε4 status, and the time interval between baseline and the tau PET scan.We observed a significant interaction between FHS-CVD and Aβ burden on subsequently measured ITC tau (p < 0.001), whereby combined higher FHS-CVD and elevated Aβ burden was associated with increased tau. The interaction was not significant for EC tau (p = 0.16).Elevated vascular risk may influence tau burden when coupled with high Aβ burden. These results suggest a potential link between vascular risk and tau pathology in preclinical AD. Ann Neurol 2019; 1-8 ANN NEUROL 2019;85:272-279.
View details for DOI 10.1002/ana.25399
View details for PubMedID 30565287
View details for PubMedCentralID PMC6351182
Interactive Associations of Vascular Risk and β-Amyloid Burden With Cognitive Decline in Clinically Normal Elderly Individuals: Findings From the Harvard Aging Brain Study.
2018; 75 (9): 1124–31
Identifying asymptomatic individuals at high risk of impending cognitive decline because of Alzheimer disease is crucial for successful prevention of dementia. Vascular risk and β-amyloid (Aβ) pathology commonly co-occur in older adults and are significant causes of cognitive impairment.To determine whether vascular risk and Aβ burden act additively or synergistically to promote cognitive decline in clinically normal older adults; and, secondarily, to evaluate the unique influence of vascular risk on prospective cognitive decline beyond that of commonly used imaging biomarkers, including Aβ burden, hippocampal volume, fludeoxyglucose F18-labeled (FDG) positron emission tomography (PET), and white matter hyperintensities, a marker of cerebrovascular disease.In this longitudinal observational study, we examined clinically normal older adults from the Harvard Aging Brain Study. Participants were required to have baseline imaging data (FDG-PET, Aβ-PET, and magnetic resonance imaging), baseline medical data to quantify vascular risk, and at least 1 follow-up neuropsychological visit. Data collection began in 2010 and is ongoing. Data analysis was performed on data collected between 2010 and 2017.Vascular risk was quantified using the Framingham Heart Study general cardiovascular disease (FHS-CVD) risk score. We measured Aβ burden with Pittsburgh Compound-B PET. Cognition was measured annually with the Preclinical Alzheimer Cognitive Composite. Models were corrected for baseline age, sex, years of education, and apolipoprotein E ε4 status.Of the 223 participants, 130 (58.3%) were women. The mean (SD) age was 73.7 (6.0) years, and the mean (SD) follow-up time was 3.7 (1.2) years. Faster cognitive decline was associated with both a higher FHS-CVD risk score (β = -0.064; 95% CI, -0.094 to -0.033; P < .001) and higher Aβ burden (β = -0.058; 95% CI, -0.079 to -0.037; P < .001). The interaction of the FHS-CVD risk score and Aβ burden with time was significant (β = -0.040, 95% CI, -0.062 to -0.018; P < .001), suggesting a synergistic effect. The FHS-CVD risk score remained robustly associated with prospective cognitive decline (β = -0.055; 95% CI, -0.086 to -0.024; P < .001), even after adjustment for Aβ burden, hippocampal volume, FDG-PET uptake, and white matter hyperintensities.In this study, vascular risk was associated with prospective cognitive decline in clinically normal older adults, both alone and synergistically with Aβ burden. Vascular risk may complement imaging biomarkers in assessing risk of prospective cognitive decline in preclinical Alzheimer disease.
View details for DOI 10.1001/jamaneurol.2018.1123
View details for PubMedID 29799986
View details for PubMedCentralID PMC6143121