Date of Award

January 2023

Document Type


Degree Name

Medical Doctor (MD)



First Advisor

Sam Payabvash


The field of radiomics provides a quantitative approach to medical imaging, wherein a large number of features extracted from medical images are used to detect clinically relevant information that is invisible to the naked eye. This increasingly prevalentmethodology has facilitated improvements in the speed and efficacy of clinical decision-making across specialties. One area where radiomics can make a particular impact is acute stroke imaging. As “time is brain” in acute stroke triage, automated tools for patient evaluation have significant potential to improve outcomes by providing timely, objective data. The present work aimed to create radiomics models from admission CT angiographic (CTA) images to quantify two variables of clinical interest for acute, anterior circulation large vessel occlusion (LVO) stroke patients: 1) predicted functional outcome, and 2) degree of collateral arterial flow.

To create our models, we automatically extracted 1116 radiomics features from the anterior circulation territory on admission CTAs of 829 acute LVO stroke patients who underwent endovascular thrombectomy (ET) in two academic centers. We trained, optimized, validated, and compared different machine-learning models to predict 1) favorable outcome (modified Rankin Scale<=2) at discharge, 2) favorable outcome at 3-month follow-up, and 3) collateral arterial flow (Miteff scale: good, moderate, or poor). Consensus scores from three neuroradiologists provided ground truth for collateral status models, and reliability of collateral status scoring was assessed by considering the concordance between each neuroradiologist’s scores.

For discharge outcome prediction, models were optimized/trained on n=494 and tested on an independent cohort of n=100 patients from Yale. Receiver operating characteristic analysis of the independent cohort showed no significant difference between best-performing Combined input models (Radiomics+Clinical+Treatment input data; area under the curve (AUC)=0.77) vs Radiomics+Treatment (AUC=0.78, p=0.78), Radiomics alone (AUC=0.78, p=0.55), or Clinical+Treatment (no radiomics data; AUC=0.77, p=0.87) models. For 3-month outcome prediction, models were optimized/trained on n=373 patients and tested on an independent cohort from Yale (n=72), and an external cohort from Geisinger Medical Center (n=232). In the independent cohort, there was no significant difference between Combined input models (AUC=0.76) vsRadiomics+Treatment (AUC=0.72, p=0.39), Radiomics (AUC=0.72, p=0.39), or Clinical+Treatment (AUC=76, p=0.90) models; however, in the external cohort, the Combined model (AUC=0.74) outperformed Radiomics+Treatment (AUC=0.66, p<0.001) and Radiomics (AUC=0.68, p=0.005) models for 3-month prediction. Collateral status models were optimized/trained on n=497 and tested on an independent cohort of n=101 patients from Yale. Radiomics models demonstrated general success in predicting the training measure of collateral status consensus score (independent validation AUC=0.77-0.78). Interrater concordance for collateral status ranged from kappa=0.02 to 0.38, with an overall kappa=0.12 (poor, p<0.005) between the three raters. Individual neuroradiologist collateral status correlation with clinical outcome ranged from r=0.09 (p=0.37) to r=0.34 (p<0.005).

Our work suggests that machine-learning signatures of admission CTA radiomics can predict outcome in acute LVO stroke candidates for ET. Such objective and time-sensitive risk stratification can guide treatment decisions and facilitate tele-strokeassessment of patients. Poor interrater correlation of collateral status among neuroradiologists limits clinical utility of collateral status models and depicts limited reliability of subjective collateral status, especially when used as a marker to determine ET eligibility or for other treatment decision-making in acute LVO stroke.


This thesis is restricted to Yale network users only. It will be made publicly available on 07/24/2024