Advanced systems for detecting biomarkers (molecules such as DNA and proteins that indicate the presence of disease) are critical for real-time diagnostic and disease monitoring devices.
Holger Schmidt, a distinguished professor of electrical and computer engineering at UC Santa Cruz, and his group have long focused on developing a unique, highly sensitive device called an optofluidic chip for detecting biomarkers.
Schmidt graduate student Vahid Ganjalizadeh led an effort to use machine learning to enhance the system by improving its ability to accurately classify biomarkers. Deep His neural network, which he developed, classifies particle signals in real time with an accuracy of 99.8%. The system is relatively inexpensive and portable for point-of-care applications. Nature Scientific Report.
When bringing biomarker detectors into point-of-care environments such as the field or clinic, the signal received by the sensor may not be as high quality as in a lab or controlled environment. This may be due to a number of factors, including the need to use cheaper chips to keep costs down, and environmental characteristics such as temperature and humidity.
To address the weak signal challenge, Schmidt and his team developed a deep neural network that can reliably identify the source of the weak signal. Using known training signals, researchers trained a neural network to learn to recognize potential variations, recognize patterns, and identify new signals with very high accuracy.
First, a parallel cluster wavelet analysis (PCWA) approach designed in Schmidt’s lab detects the presence of the signal. Neural networks then process potentially weak or noisy signals to identify their sources. The system works in real time, so users receive instant results.
“It’s all about making the most of a potentially poor quality signal and doing it very quickly and efficiently,” says Schmidt.
A smaller version of the neural network model can be run on portable devices. In this paper, researchers run the system on a Google Coral Dev board, a relatively inexpensive edge device for accelerating the execution of artificial intelligence algorithms. This means that the system also requires less power to do its work compared to other technologies.
“It proves that even compact, portable, relatively inexpensive devices can do the job for us, unlike some studies that need to run on supercomputers to make highly accurate detections.” “This makes it available for point-of-care applications, viable and portable.”
The entire system is designed to be used entirely locally, so data processing can be done without internet access, unlike other systems that rely on cloud computing. This also provides data security benefits as results can be generated without the need to share data with a cloud server provider.
It’s also designed for viewing results on mobile devices, eliminating the need to bring your laptop into the field.
“You can build a more robust system that you can take to resource-scarce or underdeveloped areas and still work,” says Schmidt.
This improved system also works with other biomarkers that Schmidt’s lab systems have been used to detect in the past, including biomarkers for COVID-19, Ebola, influenza, and cancer. Although currently focused on medical applications, the system could potentially be adapted to detect any type of signal.
To push the technology even further, Schmidt and his lab members plan to add more dynamic signal processing capabilities to the device. This simplifies the system and combines the processing techniques necessary to detect signals at both low and high concentrations of molecules. The team is also working on incorporating individual parts of the setup into the integrated design of the optofluidic chip.
