Jefferson AI Lab
Coming Together for Science
When discussion of an AI Lab began at our design group, I could not have been more excited. Ultimately, five of us took on the challenge of defining our organization's AI strategy and executing on projects which would both bring value and elevate our capabilities.
Our team delivered solutions and publications due not just to hard work, but to our diversity of training and backgrounds. The complete lack of overlap in our education and prior experience meant an extreme learning mindset, and time dedicated to uncovering what we didn't know that we didn't know. As a result, we delivered novel software, Continuing Medical Education lectures, articles, business partnerships, and grants.
Parametrizing Epilepsy Surgery
The neurosurgery department approached us with a request that would both save them time and democratize diagnostics. We were tasked with automating the process of finding the location in the brain of seizures for epileptic patients who needed surgery.
To succeed, we needed a means to clean and pre-process medical records, programmatically interpreting many kinds of imagery, and to train a model to predict from these data where to conduct surgeries. In parallel with this development effort, I immersed myself in the workflows of neurosurgery teams, designing an interface and process to make the output of our efforts continuous with their existing processes.
The Revolving Door of Stroke Readmissions
Once a patient is admitted to the hospital, it is in everyone’s best interest that they get better, go home, and stay there. Unfortunately, this last step cannot always be a reality, and some patients are readmitted in short order. Among all readmitted patients, those with stroke were identified as a high priority group. However, readmission has multiple causes within and between patients. Staff struggled to take proactive steps toward reducing readmission, both due to high workloads and opaque patterns.
Ingesting data from multiple streams across the health system, our team built a tool with over 96% accuracy and very few false positives. By presenting live readmission risk scores and highlighting contributing factors, we equipped the stroke team to not only see risks far down the road, but understand their causes, reducing stroke readmissions.
Education for Collaboration
To further engagement with our team and familiarity with use cases for AI, we designed and taught Continuing Medical Education courses on AI in healthcare. We took on conversations and education with multiple physician departments, teaching basic concepts, case studies, and deep-dives into specific topic areas, such as NLP or CV. The community we built was essential not only to growing awareness of AI, but led directly to projects with some of the physicians we taught.
Social Engagement
To educate and create community outside of Jefferson, our team published on Instagram weekly. We developed a series termed Machine Learning Mondays, where we would choose one topic or concept in AI/ML to make more accessible and tangible. This effort more than doubled engagement with our design group’s Instagram account, and started many conversations in the office and beyond, including invitations to publish our work, which we took up.