AI learns to pinpoint cancer cells by their pH level, promises early cancer diagnosis

A novel artificial intelligence (AI) system helps distinguish cancerous from healthy cells based on their acidity alone, paving the way for rapid, noninvasive single-cell classification that can facilitate early cancer diagnosis.

Developed by a team of investigators from the National University of Singapore (NUS), the system combines bright-field colorimetric pH imaging and machine learning (ML)-based image analysis. Each cancer test can be completed within 35 minutes, and single cells can be classified with an accuracy of more than 95 percent.

In a study, the AI system successfully classified nontumourigenic breast cancer cells (MCF-10A), metastatic breast cancer cells (MDA-MB-231), pancreatic cancer cells (MiaPaCa-2), and human umbilical vein endothelial cells (HUVECs). [APL Bioeng 2021;doi:10.1063/5.0031615]

“Our method was further validated using ‘in silico co-cultures’ datasets and in actual co-cultures, where a classification accuracy of 78 percent was achieved in the case of MCF-10A and MDA-MB-231 and 77 percent in the case of two cell lines that are expected to exhibit similar pH values as they are both metastatic breast cancer cell lines from the same site (MCF-7 and MDA-MB-231),” according to lead investigator Prof Chwee Teck Lim, Director of the Institute for Health Innovation & Technology (iHealthtech) at NUS.

Colour signatures

The NUS team's method utilizes a pH-sensitive dye, bromothymol blue (BTB), which changes colour according to the level of acidity of living cells. Each type of cell has its own colour ‘signature,’ which consists of a unique combination of red, green, and blue (RGB) components, owing to their distinct functions and metabolism. Compared with healthy cells, cancer cells exhibit an altered pH, and, as such, they react differently to the dye.

“The pH imaging principle is based on a recent work by Hou [and colleagues] where for the first time, live-cell imaging and single-cell intracellular pH sensing and profiling were achieved and used for cancer cell identification,” Lim said. [Sci Rep 2017;7:1-8]

He added though that they had to conduct further experiments to identify the range of ethanol concentrations to enable dye internalization without affecting the cellular physiology. The concentration of 1 mg/ml previously reported by Hou and colleagues was not well tolerated by all the cells included in the current experiments. The best trade-off in terms of BTB internalization and cell viability was 0.5 mg/ml.

Lim and his team then developed AI-based algorithm to quantitatively map the unique acidic signatures so that the cell types examined can be easily and accurately identified.

Keeping cells alive

The promise of looking at intracellular acidity to examine single cells lies in keeping the cell alive. This is a huge advantage, as current techniques to visualize specific biological phenomena occurring at the cellular and subcellular levels, such as immunofluorescence, can induce phototoxic effects or even kill the cells. 

“Unlike other cell analysis techniques, our approach uses simple, inexpensive equipment, and does not require lengthy preparation and sophisticated devices. Using AI, we are able to screen cells faster and accurately,” Lim said in a press statement. “Furthermore, we can monitor and analyse living cells without causing any toxicity to the cells or the need to kill them. This would allow for further downstream analysis that may require live cells.”

The next step for the NUS team is to expand the function of the AI system to allow detection of different stages of malignancies from the cells tested. Lim also said they are looking to develop a real-time version of their technique where cancer cells can be automatically identified and immediately isolated for further downstream molecular analysis, such as genetic sequencing, to determine any possible drug treatable mutation.

“We are exploring the possibility of performing the real-time analysis on circulating cancer cells suspended in blood,” he said. “One potential application for this would be in liquid biopsy where tumour cells that escaped from a primary tumour can be isolated in a minimally invasive fashion from bodily fluids such as blood.”