Inspired by the organization of biological neurons and the functioning of the human brain, neural networks are algorithms that include natural computing. This field is based on observation of natural phenomena to create artificial intelligence.
In order to understand the relationship between artificial and biological neural networks, it is essential to explore the human brain and understand how its numerous processes inspire the development of advanced artificial intelligence algorithms, resulting in robust and adaptable solutions.
As one of the most fascinating organs of the human body, the brain is part of the central nervous system. It contains more than 86 billion neurons that process and transmit information, controlling every function of the body and acting as the center of intelligence and learning.
This continuous evolution system, which has been functioning for millions of years, enables the brain to develop highly efficient mechanisms for processing information, learning from experiences, and adapting to new situations.
Studying and replicating these brain mechanisms is indispensable for developing AI, especially in fields like machine learning and artificial neural networks. This deeper understanding provides valuable insights into foundational concepts such as neuroplasticity and learning.
It is now widely accepted that artificial neural networks function as the “brain” of technology. This aspect of natural computing is one of the most powerful and versatile tools for artificial intelligence, and it is used in numerous fields, including the health sector, due to its ability to solve a variety of complex problems.
Neural Networks and Perception Layers
Within the concept of neural networks, there is a subfield known as “perception layers”. Human intelligence is not based on a single neuron but on the connections and coordination of many neurons working together. These neurons are organized into layers.
Inspired by the biological nervous system, an artificial neural network is essentially a computational tool made up of interconnected units, known as perceptrons or artificial neurons, which are also organized into layers. These layers are responsible for processing and transforming input data to generate the desired outputs.
In summary, the input layers gather information, the intermediate layers process the data, and the output layers handle the final actions.
How Does Biologix Utilize Neural Networks?
Biologix uses neural networks within its algorithm for snore detection. The Biologix solution was designed in 2017 and deployed in 2019 exclusively for the Biologix Sleep Test®. In 2024, the technology underwent an evolutionary process, leading to the development of the Biologix Snoring Test®, which can now be conducted without the need for the Oxistar® sensor.
Following the principle of perception layers, the learning process of a neural network is interactive, involving the introduction of repeated training data to the network. Artificial neural networks are a set of mathematical algorithms that can learn from features, identify patterns within a dataset, and classify those events.
Much like how a person learns to classify objects based on their characteristics, artificial neural networks undergo various training/learning processes:
- Training: The AI is taught to identify events based on established characteristics (in the case of snoring: volume, frequency, duration, etc.).
- Testing: The AI is challenged to apply the acquired knowledge to classify new, previously unseen events correctly.
In this process, the Biologix solution provided the neural network with characteristics that best describe snoring (e.g., a deep and persistent sound, with an average duration of 0.6 to 2 seconds, etc.) and presented numerous examples. At this stage, audio recordings from 268 patients, captured during polysomnography tests conducted at the Sleep Lab of the Heart Institute of the University of São Paulo Medical School, were used. A human evaluator listened and manually classified each recording as “snoring” or “noise”.
The neural network was trained using these pre-classified examples and features of interest. In this way, its training and learning process was initiated. It is important to note that, during this process, the network calculates a weight for each attribute, increasing the weight of those that improve performance.
In the next phase, the neural network was tested with new samples to apply its learned knowledge to classify sound events correctly.Upon completing this entire process, the tool’s performance was evaluated using different statistical criteria:
- Accuracy.
- Sensitivity.
- Specificity.
The higher these metrics, the more reliable the neural network's classification, and vice versa.
The tool implemented in the Biologix solution has already undergone various tests and improvements. These were conducted by a dedicated research and innovation team. After several iterations, the accuracy in identifying and classifying snoring is now at 92%, validated by two scientific publications: Clinics (11/2020) and Scientific Reports – Nature (09/2022).
Since this process is fully automated, it is worth noting that the privacy of the audio recordings is guaranteed for the person who undergoes the Biologix Sleep Test®. Only the responsible physician and the patient have access to this information.
