Hyderabad: A study by researchers at BITS Pilani-Hyderabad has shown that machine learning can help improve the production of “lab-on-chip” devices used in medical testing, potentially making them faster to manufacture and more reliable for healthcare use. The study, published in ‘Microsystem Technologies’, was carried out by the MEMS, Microfluidics and Nanoelectronics (MMNE) Lab at the campus.
Lab-on-chip devices are miniature systems used for chemical and biological testing with very small fluid samples. They are increasingly used in point-of-care diagnostics, biosensors, environmental testing and healthcare screening. The study focused on microchannels, the tiny pathways inside these devices that control fluid movement. Their width, depth and smoothness directly affect how accurately a test performs.
Prof. Sanket Goel said selecting fabrication settings had traditionally been “one of the most time-consuming parts” of building such devices because researchers often depend on repeated trial-and-error experiments. He said the new finding “shows is that with a reasonably sized experimental dataset, machine learning can predict channel geometry accurately enough to skip much of that trial-and-error step.”
“For groups building point-of-care or biosensing platforms, that translates directly into less wasted material and faster iteration,” Prof. Goel said. The team fabricated PMMA polymer microchannels using different carbon dioxide laser settings and analysed them through optical microscopy and profilometry before training six machine learning models on the data.
Prof. Satish Kumar Dubey said the study included both raster and vector laser modes along with pulses-per-inch settings because earlier studies had not examined them together. “The results suggest that mode of operation has a real effect on channel quality,” he said.
Prof Arshad Javed added that ensemble and boosting models performed far better than traditional linear regression methods. “Gradient boosting in particular gave us prediction errors low enough to use the model as a practical design tool,” he said, while Amit Kumar Bhagat said the models could help reduce repetitive physical experiments and improve reliability in device manufacturing.
The team said the method could support quicker and potentially cheaper production of portable diagnostic devices, especially in places with limited laboratory infrastructure.