Researchers from the European Institute of Bioinformatics, the Wellcome Sanger Institute and other institutions and their collaborators have developed an artificial intelligence algorithm that utilizes computer vision to analyze tissue samples from cancer patients. They have demonstrated that the algorithm can distinguish between healthy and cancerous tissues and can also identify patterns of more than 160 DNA and thousands of RNA changes in tumors. Related papers were recently published in the Journal of Nature-Cancer, highlighting the potential of artificial intelligence in improving cancer diagnosis, prognosis, and treatment.
The diagnosis and prognosis of cancer are mainly based on two methods. One is that a histopathologist examines the appearance of cancer tissue under a microscope. The other is cancer geneticists that analyze changes in the genetic code of cancer cells. Both methods are essential for understanding and treating cancer, but they are rarely used at the same time.
"Clinicians have always used microscope slides for cancer diagnosis. With the advancement of computer vision technology, we can analyze digital images of these slides to understand what is happening at the molecular level." Yu Fu, a postdoctoral fellow at the European Institute of Bioinformatics said.
As a kind of artificial intelligence, computer vision algorithms can recognize certain features in images. Researchers initially used this algorithm developed by Google to classify daily objects such as lemons and sunglasses, and then applied it to distinguish different types of cancer and healthy tissues. They found that this algorithm can also be used to predict the pattern of DNA and RNA changes in tumor tissue images, and even predict survival.
In total, their algorithm can detect 167 different mutation patterns and thousands of gene activity changes. These findings specifically explain how genetic mutations change the appearance of tumor cells and tissues. Another research team also used similar artificial intelligence algorithms to independently verify images of eight cancers. Related papers were also published in the journal of Nature-Cancer.
The researchers said that the combination of molecular and histopathological data provides a clearer picture of the tumor. Detecting the molecular characteristics, cell composition and survival rate associated with a single tumor will help clinicians tailor appropriate treatments according to the needs of patients.