About
This is the capstone project for the MIDS program at UC Berkeley. The goal of this project is to develop a machine learning model that can predict the likelihood of a patient having a certain disease based on various textual, tabular, and visual data.
In emergency care, healthcare workers face significant challenges due to the fragmented and varied formats of patient data. Specifically, physicians struggle with the limited access to the complex, interwoven relationships between prior doctors' notes, current vitals, and chest X-ray images. These data integration issues often result in extended wait times for diagnosis, increased healthcare costs, and delays in treatment, all of which collectively heighten patient risks.
Untapped Potential of EHR Systems
Despite the detailed patient profiles created by Electronic Health Record (EHR) systems, which include diagnostic results, radiology studies, and clinical notes, the full potential of these records for personalized patient care remains largely untapped. Current AI diagnostic tools are unable to fully leverage the diverse data types available, leading to significant gaps in the ability of healthcare providers to analyze and understand the complexities of healthcare data. This oversight impedes the advancement of personalized medicine and comprehensive patient care.
MedFusion Analytics Model
MedFusion Analytics introduces an advanced early fusion multimodal model that seamlessly integrates and interprets diverse medical data. Our innovative approach enhances the accuracy and efficiency of diagnostic processes, reduces operational costs, and improves patient outcomes. By transforming and merging insights to reveal complex relationships in EHR data previously impossible to uncover, MedFusion Analytics is reshaping the future of medical diagnostics.
As a pioneer in multimodal medical diagnostics, MedFusion Analytics accurately predicts pathological findings in chest X-rays with precision superior to individual models. Our unique combination of patient data, clinician notes, and images enables physicians to deliver more nuanced and accurate diagnoses. Tailored specifically for healthcare researchers and attending physicians, our solution facilitates advanced pathology predictions and enhances diagnostic insights, thereby elevating the quality of patient care and ensuring timely, accurate treatment decisions.
Project Repository
The code for this project can be found on our GitHub repository (opens in a new tab). The specific code for this site can be found in the demo site GitHub repository (opens in a new tab)
Capstone Project Presentation.
Specific details about this project can also be found on the UC Berkeley School of Information Project Page (opens in a new tab). You can also view our project presentation here (opens in a new tab).
Team
If you are interested in learning more about this project, feel free to contact us at: