Researchers from the University of Washington and the University of California at San Diego have developed a new
The researchers used convolutional neuralnetwork to train AI to determine blood oxygen levels without special instruments. To measure oxygenation, simply turn on the flash and camera and place your finger on them.
With every heartbeat, blood flows through the partilluminated by a flash, scientists explain. The camera records how much blood absorbs flash light in each of the three measured color channels: red, green, and blue. The data collected is used to train the AI and measure oxygen levels.
Experimental design. Image: Jason S. Hoffman et al., Digital Medicine
Each participant inhaled a controlled mixture of oxygen and nitrogen to slowly lower their oxygen levels. In total, the researchers obtained more than 10,000 readings of blood oxygen levels ranging from 61% to 100%.
The researchers tested the model onvolunteers. For each of them, oxygen levels were measured using a pulse oximeter and a new app. The results showed that the app measures oxygen saturation data in the clinical range of 70 to 100% (like most pulse oximeters), with over 80% accuracy.
The researchers note that the preliminary model was trained on a small number of volunteers, with an increase in the amount of data, the accuracy of measurements should increase even more.
Read more:
Look at a phenomenon that is simply impossible on Mars
A huge satellite can outshine all the stars and planets in the night sky
China says they will have fusion power in 6 years