← Latest papers
💻 computer science

Explainable Automated Pain Level Classification Using a Deep Neural Network with EEG Signals

This paper presents PainNet-ST, a novel explainable deep learning model that converts raw EEG signals into multi-spectral topography maps to achieve state-of-the-art 92.28% accuracy in three-class pain severity classification while enhancing clinical trust through dual-layer XAI visualizations.

Original authors: Mohammed Nafizul Islam Chowdhury Nayan, SALVIN PRASAD, RAVINESH DEO, RAJENDRA ACHARYA, PRABAL BARUA, ARUNA DEVI, JEFFREY SOAR

Published 2026-11-16
📖 4 min read☕ Coffee break read

Original authors: Mohammed Nafizul Islam Chowdhury Nayan, SALVIN PRASAD, RAVINESH DEO, RAJENDRA ACHARYA, PRABAL BARUA, ARUNA DEVI, JEFFREY SOAR

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Pain is a deeply personal experience, a warning signal that tells the body something is wrong. For doctors and nurses, measuring this signal has long relied on asking the patient to describe it, often using a simple scale from zero to ten. While this works well for many, it fails completely for those who cannot speak, such as infants, patients in intensive care, or individuals with dementia. In these cases, pain is a mystery, leading to treatments that are either too weak or unnecessarily strong. Scientists have long hoped that the brain itself holds the answer. By placing sensors on the scalp to listen to the electrical chatter of the brain, researchers believe they can find an objective measure of pain, one that does not depend on words. However, the brain's signals are incredibly complex and noisy, making it difficult to separate the specific pattern of pain from the background noise of everyday thought and movement.

A team of researchers has now developed a new way to listen to these brain signals, creating a system that can automatically tell the difference between low, medium, and high levels of pain. They built a computer program that acts like a highly trained observer, capable of spotting subtle patterns in brain activity that humans would miss. Instead of asking a patient to rate their pain, this system looks at the raw electrical waves recorded from the scalp. The researchers tested their method on data from 223 healthy volunteers who had received small, controlled laser pulses to their hands. These pulses were designed to trigger a pain response without causing actual injury. The goal was to see if the computer could correctly identify how intense the pain felt based solely on the brain's reaction to the laser.

The system they created, which they named PainNet-ST, works by looking at the brain's activity in two ways at once: across the surface of the head and over time. Imagine the brain as a vast landscape where different areas light up when pain strikes. The computer first maps out which parts of the scalp are most active, focusing on the central and back regions where pain signals are strongest. It then watches how these signals change in the milliseconds following the laser pulse, looking for a specific wave pattern that rises and falls in a predictable rhythm. This dual approach allows the system to understand both where the pain is being processed and how it evolves moment by moment. To ensure the system is not just memorizing the specific brains of the people it was trained on, the researchers tested it on completely new people it had never seen before. This is a crucial step, as it proves the system can recognize pain in anyone, not just in a specific group of volunteers.

The results were striking. When the computer analyzed the brain signals, it correctly identified the level of pain in 92.28% of the cases. This is a significant improvement over previous methods, which often struggled to distinguish between different intensities of pain or failed when tested on new people. The system was particularly good at spotting high levels of pain, which is vital for ensuring that patients in critical condition receive the help they need. It also made very few mistakes in identifying when pain was not present, meaning it would not falsely alarm doctors with unnecessary treatment. To make sure the system was not a "black box" that made decisions without reason, the researchers added a layer of transparency. They built a feature that highlights exactly which parts of the brain and which moments in time the computer used to make its decision. This confirmed that the system was indeed looking at the known biological markers of pain, giving doctors confidence that the results are trustworthy and grounded in real science.

This work represents a major step forward in making pain assessment objective and reliable. By proving that a computer can learn to recognize pain from brain signals alone, the researchers have opened the door for tools that could one day help those who cannot speak for themselves. While the current system was tested in a controlled setting with healthy volunteers, the findings suggest that this technology could eventually be adapted for real-world hospitals. The ability to measure pain without words could transform how medical teams care for vulnerable patients, ensuring that pain is treated with the same precision and care as any other medical condition. The study shows that with the right tools, the hidden language of the brain can be translated into clear, actionable information for healing.

Drowning in papers in your field?

Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.

Try Digest →