Assistant Professor Yu-Chih Chen is harnessing the power of computational algorithms to improve cancer research.
His latest article, “Deep Learning Unlocks Label-Free Viability Assessment of Cancer Spheroids in Microfluidics,” was published in Lab Chip.
“In this data science era, I want to use microfluidic tools, image processing and a label-free approach to get more data,” Chen said. “Improved data collection has the potential to revolutionize the data discovery process.”
Microfluidics refers to devices that manipulate fluids and cells within small channels and chambers. In his research, Chen utilizes semiconductor manufacturing capability to create chambers that grow uniform cancer spheroids. Spheroids are three-dimensional (3D) cell cultures, which better mimic the behavior of tumors.
The 3D model represents an improvement over the traditional 2D cultures because they are more realistic and can be designed to resemble the patient’s situation.
“The goal is to build a reliable method for forming and characterizing numerous 3D tumoroids,” he said. “In the future, when we do high throughout cancer spheroid experiments, we can quickly test many spheroids with thousands of drugs or drug combinations.”
To test his method, Chen trained a deep learning model with eight conventional chemotherapeutic drugs. The model then predicts the results generated by other compounds or lab images.
“It’s a cycle,” Chen said. “Once we know the ground truth, we can compare the prediction versus the ground truth to know whether this model can really predict what’s going on.”
A key innovation of this study is the application of computational methods to label-free dynamics. Label-free detection uses phase contrast microscopy, a method to observe cells through phase shifts in light. It provides a high-throughput and non-destructive way to watch the life cycle of a tumor spheroid in real time without fluorescent dyes that can interfere with cell behavior. This viability dynamic helps scientists understand how cancer cells are killed by treatment over time.
“The part of I’m proud of for this work is we did really comprehensive characterization,” he said. “So, we tested different drugs, different cell lines, and even images from other labs and got consistently valid predictions. We envision our work can be widely used for other cell types, drugs and images collected in other labs.”
In the future, this model will support high throughput screening, empowering researchers to test thousands of compounds. Better drug testing practices support the end goal of finding effective treatments for cancer, improving the lives of patients everywhere.