Related Experiment Video
Updated: Jul 7, 2025
09:41
Functional Transcranial Doppler Ultrasound for Monitoring Cerebral Blood Flow
Published on: March 15, 2021
7.9K
A data-driven computational methodology towards a pre-hospital Acute Ischaemic Stroke screening tool using
Ahmet Sen1, Laurent Navarro1, Stephane Avril1
1Mines Saint-Etienne, Univ Jean Monnet, INSERM, U 1059 Sainbiose, F-42023, Saint-Etienne, France.
Computer Methods and Programs in Biomedicine
|December 22, 2023
Summary
Machine learning models can detect and locate blood clots in Acute Ischaemic Stroke (AIS) patients using haemodynamic data from Doppler Ultrasound. This approach offers a faster, cost-effective alternative to traditional imaging for stroke diagnosis.
Area of Science:
- Biomedical Engineering
- Medical Imaging
- Machine Learning Applications in Healthcare
Background:
- Acute Ischaemic Stroke (AIS) is a leading cause of disability and mortality worldwide, necessitating rapid diagnosis and intervention.
- Current diagnostic methods like "drip and ship" rely on advanced imaging in specialized centers, causing critical delays in treatment initiation.
- Timely detection and localization of cerebral artery occlusions are crucial for effective patient management and improved outcomes.
Purpose of the Study:
- To investigate the feasibility of a machine learning model for diagnosing and locating occluding blood clots in AIS.
- To utilize velocity waveforms obtained from portable Doppler Ultrasound devices as an alternative diagnostic tool.
- To develop a cost-effective and time-efficient screening method for Acute Ischaemic Stroke.
Main Methods:
- Simulated hemodynamic data representing healthy and AIS scenarios were generated using a population-based database.
- A machine learning classification model was trained to detect and locate thrombi by analyzing measured waveforms.
- The model performed a two-step classification: identifying the affected region and classifying the thrombus size (small, medium, large vessel occlusion).
Main Results:
- The methodology achieved over 95% true prediction rate for both classification steps in noise-free simulated data.
- With up to 20% noise, the true prediction rate decreased to 80% for region detection and 70% for bifurcation generation detection.
- These results indicate high accuracy in identifying clot location and type under ideal conditions.
Conclusions:
- The study demonstrates the potential of using hemodynamic data and machine learning for efficient detection and localization of thrombi in AIS.
- The proposed approach shows promise as a viable, rapid, and accessible alternative to conventional imaging techniques.
- Further adjustments are needed for real-world application, but the idealized data results are encouraging for clinical translation.
