For real-time diagnostic and disease-monitoring devices, sophisticated systems for the detection of biomarkers—molecules like DNA or proteins that indicate the presence of a disease—are essential.
UC Santa Cruz’s distinguished electrical and computer engineering professor Holger Schmidt and his group have long focused on creating optofluidic chips, which are one-of-a-kind biomarker detection devices that are extremely sensitive.
Vahid Ganjalizadeh, a graduate student of Schmidt’s, was in charge of a project to use machine learning to improve their systems’ ability to accurately classify biomarkers. According to a new Nature Scientific Reports paper, the deep neural network he created classifies particle signals in real time with 99.8 percent accuracy on a system that is relatively inexpensive and portable for point-of-care applications.
While taking biomarker finders into the field or a purpose in care setting, for example, a wellbeing center, the signs got by the sensors may not be essentially as excellent as those in a lab or a controlled climate. This could be because of a number of things, like the need to use cheaper chips to cut costs or the temperature and humidity of the environment.
A deep neural network that is capable of identifying the source of a weak signal with high confidence was developed by Schmidt and his team to address the difficulties posed by weak signals. In order for the neural network to be able to recognize patterns and identify new signals with extremely high accuracy, the researchers taught it to recognize potential variations by training it with known training signals.
First, a method called parallel cluster wavelet analysis (PCWA) that was developed in Schmidt’s lab finds that there is a signal. The neural network then identifies the signal’s source after processing the potentially weak or noisy signal. Since this system works in real time, users can get their results right away.
Schmidt stated, “It’s all about making the most of possibly low quality signals in a really quick and efficient manner.”
A more modest variant of the brain network model can run on convenient gadgets. In the paper, the system is run on a Google Coral Dev board, a relatively inexpensive edge device for accelerating the execution of AI algorithms. As a result, the system uses less power to carry out the processing than other methods.
“We proved that even a compact, portable, and relatively cheap device can do the job for us,” Ganjalizadeh said. “In contrast to some research that requires running on supercomputers to do high-accuracy detection.” It makes it feasible, portable, and available for use at the point of care.”
The whole framework is intended to be utilized totally locally, meaning the information handling can occur without web access, dissimilar to different frameworks that depend on distributed computing. Because results can be produced without sharing data with a cloud server provider, this also improves data security.
Additionally, it is made to deliver results on a mobile device, removing the requirement to bring a laptop into the field.
Schmidt stated, “You can build a more robust system that you could take out to less developed or under-resourced regions, and it still works.”
Any additional biomarkers that Schmidt’s lab’s systems have previously been used to detect, including COVID-19, Ebola, flu, and cancer biomarkers, will work with this improved system. In spite of the fact that they are presently centered around clinical applications, the framework might actually be adjusted for the discovery of a sign.
To drive the innovation further, Schmidt and his lab individuals intend to add considerably more powerful sign handling capacities to their gadgets. The system will be made simpler and the processing methods needed to detect signals at low and high molecule concentrations will be combined as a result. The group is likewise attempting to bring discrete pieces of the arrangement into the incorporated plan of the optofluidic chip.