Detecting cancer: Artificial Intelligence beats humans

Editorial

Rebecca Pool

Tuesday, February 5, 2019 - 10:15
Phase contrast images of cancer cells and radioresistant cells [Osaka University]
 
Artificial intelligence can identify different types of cancer cells better than humans, claim Japan-based researchers.
 
By applying deep learning to phase contrast microscopy images, Professor Hideshi Ishii from Osaka University and colleagues identified minute differences between, human and mouse cancer cells, as well as the radioresistant clones, to 96% accuracy.
 
This new approach to cancer cell differentiation opens the door to better cancer treatments.
 
Researchers have recently been using convolutional neural networks to analyse medical images for clinical applications.
 
However, Ishii and colleagues, wanted to see if such artificial intelligence could be trained to distinguish microscopy images of mammal cells.
 
With this in mind, they used thousands of phase-contrast microscopy images of mouse squamous cell carcinoma and human cervical carcinoma, as well as the radioresistant clones, to train a deep convolutional neural network, VGG16, typically used for large-scale image recognition.
 
“We first trained our system on 8000 images of cells obtained from a phase-contrast microscope,” highlights Ishii. “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.”
 
Analyses revealed that the neural network correctly identified 96% of the cells in their dataset, although it had greater success at recognising the mouse cells lines.
 
“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 tumour or circulating in the body of cancer patients,” says Ishii's colleague Masayasu Toratani, Osaka University.
 
“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,” adds Toratani.
 
The researchers reckon their latest results validate the use of AI identify differences in phase-contrast microscopy images and now intend to train their convolutional neural network to recognise a wider range of cancer cells.
 
They eventually hope to set up a universal AI-based system for automatically determining any kind of cancer cell.
 
Research is published in Cancer Research.
 
Website developed by S8080 Digital Media