Understanding machine learning is to modify the delivery of cold atmospheric plasma drugs in cancer treatments without being trained with detailed plasma parameters.
While artificial intelligence (AI) can adapt to changing conditions and achieve the desired results, it can become a mystery how algorithms “understand” and adjust the input.
Lin et al. I tried to discover this “black box” with AI-controlled cold atmospheric plasma (CAP) treatment. This is an approach that induces apoptosis in diseased cells while preserving healthy cells. Previous studies have developed a machine learning (ML) system that predicts the post-treatment status of cancer cell targets and tailors treatments accordingly. However, we did not know how the ML system achieved this result without understanding the specific plasma parameters.
Using an AI-based light emission spectroscopy (OES) spectral translation algorithm, the authors inverse engineered real-time chemical accumulation on the surface of cell media. They found that despite the changes in conditions, the ML algorithm changes experimental parameters to achieve the same treatment outcome. Applying the Fourier transform to OES spectra and chemical kinetic rate analysis revealed how the ML algorithm independently captured the way in which an additional layer of physical information, which relies solely on cell viability status without human input of this information, achieved the accuracy and reliability of AI-controlled CAP models.
“Beyond plasma medicine, a similar approach can advance machine learning-based control in areas such as satellite electric propulsion, plasma-based microfabrication, fusion reactor management, and many other plasma applications,” says author Michael Keidar.
The team is then trying to expand the range of control demonstrated in this paper.
“Instead of limiting AI to adjusting processing duration, we plan to approve and train AI to simultaneously control multiple plasma parameters, including voltage, gas flow and even additional external electric fields,” author Li Lin said. “In doing so, we aim to tailor treatments to each patient's specific needs.”
sauce: “Cryoplasmic adaptation in the process of machine learning control in plasma medicine,” Li Lin, Qihui Wang, Zichao Hou, Michael Keidar, Plasma physics (2025). You can access the article https://doi.org/10.1063/5.0274614 .
