MIRAGE
Team Member(s)
Ahmed Alqalam
Nicolo Krueger
Andrew Cassarino
Mentor(s)
Dr. Syed Anwar, Children's National Hospital
Nishad Kulkarni
Faisal Al Munajjed
Zhenhao Zhao
Project Sponsor(s)
Dr. Syed Anwar, Children's National Hospital
Instructor(s)
Dr. HyungSok Choe, BME, GW Engineering
Many patients presenting to hospital emergency rooms require imaging for diagnosis. Although X-ray or CT scans can be performed quickly, not having a doctor qualified to interpret the image available is a significant obstacle to positive health outcomes. MIRAGE provides nurses and medical fellows a resource to take immediate action for their patients; as a result, the potential detriment of not having a medical practitioner’s analysis at any given moment will decrease substantially.
Although we believe our approach is powerful enough to give an accurate analysis, it should be used as a first step without replacing the doctor’s analysis. As LLM technology continues to develop, this may change, but for now ensuring a patient’s health during this period of waiting is vital.
Who experiences the problem?
Two key groups experience this problem: radiologists and patients. Radiologists face increasing workloads, long hours, and pressure to quickly interpret complex imaging cases, which contributes to a high rate of burnout. This affects their ability to keep up with demand, leading to delays in image interpretation. As a result, patients are left waiting for imaging to be reviewed, which postpones critical treatment decisions.
Why is it important?
This problem is important because it directly impacts patient safety and the effectiveness of emergency medical care. In critical conditions, even short delays in diagnosis could lead to substantial complications. Therefore, providing immediate AI-supported insights into radiological imaging allows medical staff to act faster and more confidently while waiting for formal interpretation. This not only improves workflow efficiency, but also aims to empower healthcare professionals with a cutting-edge tool that combines knowledge retrieval and generation, so they can focus on their primary mission: delivering the highest standard of patient care.
What is the coolest thing about your project?
The coolest thing about our project is the ability to combine multimodal AI with real clinical use cases to assist healthcare providers. By integrating images and text inputs into our RAG framework, our system mimics the cognitive workflow of a radiologist.
What sustainable design considerations drove your solution?
Our RAG framework prioritizes sustainability by focusing on localized AI systems. Rather than relying on cloud connectivity such as OpenAI, we plan to integrate open-source models such as Llama 3.2 11B. This choice enhances data security while minimizing environmental impact.
What were some technical challenges?
One of the key technical challenges we faced was integrating our AI system into the hospital's computing environment. Our initial development of the system used the OpenAI API, which was compatible across operating systems. However, when transitioning into the Llama 3.2 11B model for local deployment, we encountered compatibility issues. It only ran effectively on a Windows OS. To solve this, we had created a Docker file with the help of Project Mentor, Nishad Kulkarni, to establish a stable environment for the hospital's machine to run MIRAGE.