Leverage electronic data collection for clinical and scientific inquiry.
The missions core is the development of a unique patient centered approach through the incorporation of objective disease outcomes, patient focused results and value based care delivery. Our aim is to increase quality and reduce cost through an efficient and responsive engagement of patients and health care providers through the use of novel methodologies and state of the art research for the transformation of clinical care.
The goal of this project is to capture moment-by-moment information regarding the wellbeing of our patients using data from personal digital devices. This allows us to capture the lived experience of patients and their interaction with the world with minimal interference in order to better understand and create outcome measures that are important to them.
Accurate labelling of brain tumors is called segmentation and allows for accurate volumetric analysis of tumors. This allows surgeons to monitor tumor growth or recurrence with high precision, and may help us predict growth patterns in the future. In this project, we use machine learning (deep neural networks) to automate the segmentation of meningiomas, the most common type of primary brain tumor.
The majority of data in the electronic medical record is contained in free-form text reports including physician notes, allied health services notes, radiology and pathology notes to name a few. Our ability to gain insight from this rich data source is limited by the need for manual review which is labor intensive, time consuming and error prone. This project aims to use machine learning (natural language processing) to automate much of the work in chart review so that researchers can instead focus on asking the questions.
Glioblastomas are the most common malignant brain tumors, and are often treated with surgery followed by chemotherapy and radiation. Not all glioblastomas are created equal, however, and molecular markers help guide the treatment plan and outlook on prognosis. Molecular markers are identified from tissues at the time of tumor removal. In this project, we use machine learning (convolutional neural networks) to predict the status of molecular markers prior to surgery to help guide patients and surgeons.
In this institutional radiographic imaging analysis, traditional methods of ventriculostomy site selection predicted significant rates of cortical vein injury, matching described rates in the literature. CTA/CTV imaging potentiates identification of patient-specific cannulation sites and custom trajectories that avoid cortical vessels, which may lessen the risk of intracranial hemorrhage during ventriculostomy placement. Further development of this software is underway to facilitate stereotactic ventriculostomy and improve outcomes. Learn more >