Intel® AI Hackathon
CardioLens - Automated Echocardiogram Segmentation along with Report Generation Using Intel Optimized Models
Team Details | |
Team Name: | Three of Hearts |
Student 1 Name: | Nikhileswara Rao Sulake |
Student 2 Name: | Sai Manikanta Eswar Machara |
Student 3 Name: | Aravind Raju Pyli |
Student 4 Name: | |
Mentor Name: | Sivalal Kethavath |
Institution Name: | Rajiv Gandhi University of Knowledge Technologies, Nuzvid Andhra Pradesh |
Submission ID: | 419 |
| |
Hackathon Theme: | HealthCare |
| |
Date: | 10-12-2024 |
Submission ID: 419
Overall error rate : 3-4%
Diagnostic error severity:
Error Preventability - 77%
Errors by Echocardiographers
Some existing softwares are charging Thousands of Dollars which is not affordable
High pricings of existing softwares
Manual testing needs more time and efforts
Time consuming
No active application of AI in Echocardiogram testing
Limited usage of AI
Current Scenario - Cardiography
Submission ID: 419
Left Ventricular Ejection Fraction
Why Left Ventrile?
Oxygenated blood
Why not Right Ventricle?
It tells about the percentage of blood the left ventricle pumps out with each contraction. A normal person heart’s ejection fraction is between 55 and 70 percent.
Most used test for EF : Echocardiogram
Submission ID: 419
CardioLens - The Solution
For Patients
For Echocardiographers and Medical Students
0.14
0.78
Dice Coefficient
Model Loss
~1.5
Inference Time
Submission ID: 419
Educational Impact
Economic Impact
Healthcare Ecosystem Impact
Market Scope of Cardio-Machinery
Impact & Influence
Submission ID: 419
AI Solution - Complete Procedure
Submission ID: 419
Intel Technology Used
Processor used : 12th-Gen-Intel(R) core(TM) I7-12650H(16 CPUs)
Clock at ~2.7 GHz, Memory : 16.3 GB
Submission ID: 419
Optimization Results
By further experimentation for inference task, we present you the results. The above figures tells about the time in seconds required to process the video for segmentation task and ejection fraction calculation task which is done by video model (R2plus1D - Video Processing). In both task we can see and confirm that using OpenVINO we can reduce the inference time by half comparing to Pytorch.
Submission ID: 419
Future Works and Scope
The above research work has been done for Adults and we would like to extend our work to analyze and detect Cardiac diseases in Infants (Babies less than 3-6 Months), we are targetting to propose a new technique to predict the Time to Positive (Days) for the detected disease which could help in further diagnosis.
Future Works