A system based on a convolutional neural network (CNN), developed at Osaka University, has shown that it is able to automatically distinguish between different types of cancer cells. Being able to accurately identify the particular cancer cell types present in a patient can be useful in choosing the most effective treatment, but current methods are time-consuming and often hampered by human error and the limits of human sight. The researchers used the CNN-based system to distinguish between mouse cells and human cells and their radioresistant clones. “We first trained our system on 8000 images of cells obtained from a phase-contrast microscope. We then tested its accuracy on another 2000 images, to see whether it had learned the features that distinguish mouse cancer cells from human ones, and radioresistant cancer cells from radiosensitive ones,” said researcher Hideshi Ishii. These are representative microscopic images of cancer cells and radioresistant cells. Courtesy of Osaka University. The features extracted by the trained CNN were plotted using t-distributed stochastic neighbor embedding. Images of each cell line were well clustered, indicating that, after training, the CNN-based system was able to correctly identify cells based on microscopic images of them. The trained CNN obtained an accuracy of 96 percent. The team’s findings suggest that image recognition using AI could be useful for predicting small differences among phase-contrast microscopic images of cancer cells and their radioresistant clones. This is the trend of the accuracy when predicting cell type based on cancer cell images. The accuracy reached about 96 percent for test data (data not used for AI training). Courtesy of Osaka University. “The automation and high accuracy with which this system can identify cells should be very useful for determining exactly which cells are present in a tumor or circulating in the body of cancer patients,” researcher Masayasu Toratani said. “For example, knowing whether or not radioresistant cells are present is vital when deciding whether radiotherapy would be effective, and the same approach can then be applied after treatment to see whether it has had the desired effect.” In the future, the team hopes to train the system on more cancer cell types, with the eventual goal of establishing a universal system that can automatically identify and distinguish all such cells. The research was published in Cancer Research (https://doi.org/10.1158/0008-5472.CAN-18-0653).