A new artificial intelligence model has identified 13 different types of cancer with an accuracy of 98.2% based on DNA data from tissue samples, according to a study.
The AI model, EMethylNET, was developed by experts at the University of Cambridge in the UK. Researchers trained the neural network to detect structures and pathways of cancer development at an early stage. The work focuses on DNA methylation—a chemical process occurring in the early stages of cell growth, including cancerous cells.
“Cancer, a collection of over 200 different diseases, remains one of the leading causes of morbidity and mortality worldwide. Metastatic cancer is typically detected at late stages of the disease, accounting for 90% of cancer-related deaths. Therefore, early detection of the disease combined with modern treatment methods will have a significant impact on the survival and treatment of various types of cancer,” the study states.
EMethylNET was trained on data from over 6,000 tissue samples from the Cancer Genome Atlas, representing 13 types of cancer. The model was tested on more than 900 samples from independent data sets.
The neural network not only accurately detects cancer but also provides insights into the regulation of non-genetic factors by the organism that mutate normal cells into cancerous ones.
The technology requires further study and testing before clinical trials, experts emphasized. Work is currently underway to adapt the model for liquid tissues.
“Depending on the availability of training data, this method could be expanded to detect hundreds of types of cancer,” the report states.
Earlier, researchers from MIT developed an AI model for determining ambiguous results in medical images.
In November 2022, Google licensed a research AI algorithm for breast cancer screening to iCAD.
In October, AI was taught to detect brain cancer from a drop of blood.