Speaker
Description
Cardiovascular disease remains the leading cause of death in the U.S, with fatalities increasing from 928,741 in 2023 to 931,578 in 2024. Moreover, there has been immense pressures on the healthcare system due to delayed or inaccurate diagnosis for heart-related issues and deaths. To bridge this gap between high heart-related deaths and low treatment plans, EASyExam is an Al-driven diagnostic tool designed to provide faster and more accurate assessments for patients. Prior teams developed the model to convert heart ultrasound videos into images under a merged framed approach, to achieve an 89.7% validation accuracy. But the model faces other challenges outside of that as well such as data limitations, spatial and temporal issues that need revision to improve overall Al accuracy and interpretability. Part of this year's research goal was to increase the model's accuracy and address the model's constraints to further aid doctor's decisions and treatment plans related to heart issues. These methodologies included creating a sliding window implementation for better data processing, a CNN-LSTM architecture for both spatial and temporal feature analysis and the GRAD-CAM overlay which displays a heatmap. Implementing different architectures and methodologies allowed us to analyze precision, accuracy and loss results from the dataset. Overall, we were able to optimize the Al-driven diagnostic tool to offer more precise and reliable heart ultrasound assessments, reducing the chances of misdiagnosis and aiding in faster clinical decision-making. These advancements pave the way for improved patient outcomes and demonstrate the potential of Al in transforming cardiovascular disease diagnostics.
| Abstract Category | Medical Physics |
|---|