Sanjana Reddy
Machine Learning Engineer, Applied AI
Research Areas
Authored Publications
Sort By
SAHVAI-3D and 4D: A New, Automated Subarachnoid Hemorrhage Volumetric Artificial Intelligence (SAHVAI) Measurement Approach Using Non-Contrast Head CT Scans
Vikas Gupta
Rabih Tawk
Yujia Wei
Quincy Gu
William D Freeman
Rohan Sharma
Vishal Patel
Saif Salman
Melina Wirtz
Stroke: Vascular and Interventional Neurology (2024)
Preview abstract
To automate subarachnoid hemorrhage volume (SAHV) calculation (SAHVAI-SAHV Artificial Intelligence) and create 3D volumetric images (SAHVAI-3D) using non-contrast head CT (NCCT) imaging data in aneurysmal subarachnoid hemorrhage (SAH) patients. We also defined SAHVAI-4D, representing SAHV over time. The aim was to compare automated SAHVAI volumes to manual SAHV methods and computation times, explore these imaging biomarkers’ potential in identifying at-risk brain regions for delayed cerebral ischemia (DCI), and explore potential insights in future neurotherapeutic interventions for SAH patient recovery.
A training set of 10 consecutive aneurysmal SAH cases was used to manually compute SAHV, SAHVAI-3D, and SAHVAI-4D, involving 92 non-contrast CT scans (182 slices each). The SAHVAI deep learning (DL) algorithm generated automated SAHV values in cubic centimeters (cc). For both SAHVAI and manual evaluations, a 3D SAH brain map was created for each patient. Blood volumetric outputs were analyzed and compared to neurological outcomes at discharge, including DCI events, symptomatic vasospasm (sVSP), and areas with the thickest SAH blood concentration.
SAHVAI quantified SAH blood volume (SAHV) in average of 6.7 seconds per scan, significantly faster than the manual method, which took over 60 minutes per scan (Fisher’s exact test, P value <0.001). SAHVAI demonstrated an accuracy of 99.8%, a Dice score of 0.701, a false positive rate of 0.0005, and a negative predictive value of 0.999. The mean absolute error between SAHVAI and manual methods was 5.67 ml. The SAHVAI-3D brain map and total SAHV at admission were strongly associated with neurological outcomes, inversely with Glasgow coma scale (R2=0.23, p=0.017) and directly with length of hospital stay (R2=0.175, p=0.004), especially in regions with dense blood concentration.
SAHVAI-3D and SAHVAI-4D brain mapping techniques represent innovative imaging biomarkers for SAH. These advancements enable rapid evaluation and targeted interventions, potentially improving patient care in SAH management.
View details
Artificial Intelligence and Machine Learning in Aneurysmal Subarachnoid Hemorrhage: Future Promises, Perils, and Practicalities
Patrick Vanderboom
Lin Lancaster
Quincy Gu
William D Freeman
Rohan Sharma
Saif Salman
Journal of the Neurological Sciences (2023)
Preview abstract
Aneurysmal subarachnoid hemorrhage (SAH) is a subtype of hemorrhagic stroke with thirty-day mortality as high as 40%. Patients develop a myriad of short and long-term complications, which peak around days four to fourteen. The use of Machine Learning (ML) and Artificial intelligence (AI) methods (such as computer vision, Convolutional Neural Networks (CNN), Natural Language Processing (NLP), and drug discovery approaches among others) are emerging in healthcare and can help a patient population desperately in need of an integrated AI system that detects, segments, and provides clinical decision support based on severity for emergency treatments using CT brain imaging and clinical informatics.
This review aims at 1) synthesizing the current state of the art of AI and ML tools available for the management of aneurysmal SAH patients, and 2) providing an up-to-date account of future horizons in patient care.
View details
Hemorrhage Evaluation and Detector System for Underserved Populations: “HEADS UP”
Patrick Vanderboom
Quincy Gu
William D Freeman
Rohan Sharma
Saif Salman
Mayo Clinic Proceedings: Digital Health (2023)
Preview abstract
Intracranial hemorrhage (ICH) is the second most common cause of stroke yet remains the second leading cause of disability globally and disproportionately deadlier than ischemic stroke. ICH also disproportionately impacts minorities and female mortality, particularly in rural and underserved areas of the globe. The noncontrast head CT (NCCT) remains the gold-standard way to differentiate ischemic stroke versus hemorrhagic stroke. Since 2015 there has been a paradigm change approach to ischemic stroke patients using AI-ML using NCCT and CT perfusion to select patients for large vessel occlusion thrombectomy to improve outcomes. ICH patients, however, lack such a paradigm change in the acute stroke systems-of-care to improve outcomes. Therefore, there is an unmet patient need to create an earlier ICH detection method to trigger accelerated systems of care that are linked to better downstream patient outcomes and interventions. The objective is to create a rapid, cloud-based, deployable machine learning (ML) method to detect ICH potentially across the Mayo Clinic enterprise to help patients and clinical teams and to create scientific and intellectual independence from third party vendors.
View details
