AI can now detect cancer cells in blood

September AI Labs have developed a method of identify circulating tumour cells in blood using computer vision and advanced machine learning prediction techniques in collaboration with Edith Cowan University.

Cancer research has taken a leap forward in quickly identifying the cancer cells that are responsible for the spread of cancer around the body. 

Our team at September AI Labs have joined forces with Edith Cowan University (ECU) Melanoma researchers to develop a way to use artificial intelligence to accurately identify cancer cells circulating in the blood.

Cancer spreads around the body when tumour cells shed from the primary tumour and travel through the blood to form secondary tumours (metastases) in other organs.

“By detecting and counting these circulating tumour cells (CTCs), clinicians and doctors can better understand what stage a cancer is at and predict the likelihood of a patient’s responsiveness to different treatments, thus improving patient outcomes” said ECU’s Associate Professor Elin Gray.

AI reduces detection time from hours to seconds

The CTCs are incredibly difficult to spot among thousands of other cells and matter in blood, they are very rare much like finding a needle in a haystack.

Within one millilitre of blood, there is often less than ten cancer cells amongst one billion red cells and one million white blood cells.

“Until now, it has taken a trained technician few hours per patient sample to manually filter different characteristics of cells using traditional imaging techniques,” Professor Gray said. “This AI technology has reduced this process down to a few minutes per patient.”

Trained in deep learning neural networks

Using more than 4,000 images from the Melanoma Research Group at ECU, the September AI team developed a machine learning model that was trained to identify circulating tumour cells with a 97 per cent accuracy.

September AI Labs Managing Director Brad Dessington said the detection of CTCs was a particularly complex challenge for machine learning to achieve such high accuracy.

“CTCs are organic biological shapes, no one is the same as each one is different in size, shape and presents in random positions among healthy cells in the blood,” Mr Dessington said.

“This was not just a matter of spotting the molecular signatures, the model had to be able to learn and understand complex images, to do it as well as a human, but far faster with robust convolutional neural networks and amped up computer power.” Says Fedja Hadzic, Chief Data Scientist at September AI.

ECU has entered into a partnership with September AI Labs to ramp up artificial intelligence and machine learning accessibility for research across the University.

Through this partnership, the CTC identification technology will also be broadened to investigate a range of other cancers including lung, breast, pancreatic and prostate.

Share now

February 24, 2020
Artificial Intelligence | Computer Vision | Medical | Neural networks
Share This