Capturing Cancer Through CTCs
Machine learning can help optimize liquid biopsy for rare circulating tumor cells
Michael Schubert | | Quick Read
What if you could offer patients with cancer all the benefits of a biopsy – without the biopsy? That compelling idea – precision medicine with no need for invasive investigations – is what makes liquid biopsy such an attractive prospect. Unfortunately, there’s a catch; circulating tumor cells (CTCs) are rare, presenting a detection challenge whose only solutions can be difficult and time-consuming. To expand the available options, one research collaboration has taken a new direction: computer assistance.
Unlike existing methods, the new approach requires no labeling or complicated microscopic techniques. Instead, a machine learning algorithm examines standard, low-resolution brightfield microscopy images to distinguish between CTCs and other cells (1). The scientists creating it ran two separate experiments: one on cultured cancer cell lines with a training set of 1,745 single-cell images; another on patient CTCs with a training set of just 95 images. The result? The algorithm exhibited an overall accuracy of 97.5 percent in the first experiment and 88 percent in the second.
“This study, though small, demonstrates that our method can achieve high accuracy on the identification of rare CTCs without the need for advanced devices or expert users,” said senior author Yaling Liu in a recent press release (2). “With more data becoming available in the future, the machine learning model can be further improved and serve as an accurate and easy-to-use tool for CTC analysis.”
So what’s next for the researchers? They’re not only training their algorithm with additional data, but also refining it to examine mutations in the DNA of CTCs. In addition, they’re working on a microfluidic device to better capture and release CTCs (3) – all with the goal of fast, accurate, and minimally invasive personalized medicine for patients with challenging cancers.