Hyderabad: A study by researchers at BITS Pilani-Hyderabad has shown that machine learning could help improve the production of “lab-on-a-chip” devices used in medical testing, making them faster to manufacture and more reliable in medical applications. The research, published in Microsystem Technologies, was conducted by the campus’ MEMS, Microfluidics, and Nanoelectronics (MMNE) laboratory.
Lab-on-a-chip devices are small systems used for chemical and biological testing using very small volumes of fluid samples. It is increasingly being used in point-of-care diagnostics, biosensors, environmental testing, health checks, etc. This study focused on microchannels, the tiny pathways within these devices that control fluid movement. Their width, depth, and smoothness directly affect the accuracy of test execution.
Professor Sanket Goel said choosing the manufacturing setup has traditionally been “one of the most time-consuming parts” of building such devices, as researchers often use trial-and-error experiments. He said the new findings “show that given a moderately sized experimental data set, machine learning can predict channel shape accurately enough to eliminate many of the trial-and-error steps.”
“For groups building point-of-care or biosensing platforms, it translates directly into less material wastage and faster iteration,” Professor Goel said. The research team fabricated PMMA polymer microchannels using different carbon dioxide laser settings and analyzed them by optical microscopy and profilometry before training six machine learning models on the data.
Professor Satish Kumar Dubey said the study included both raster and vector laser modes and pulse-per-inch settings, as previous studies had not looked at them together. “The results suggest that the mode of operation has a real impact on channel quality,” he said.
Professor Arshad Javed added that the ensemble and boosted models performed much better than traditional linear regression techniques. “Specifically with gradient boosting, the prediction error is now low enough that the model can be used as a practical design tool,” he said. Amit Kumar Bhagat said the model could help reduce repetitive physical experiments and improve reliability in device manufacturing.
The research team said this method could support faster and potentially cheaper production of portable diagnostic devices, especially in places with limited laboratory infrastructure.
