Artificial intelligence to examine biopsy tissue samples

Enginyers d'UCLA han donat amb una tinció virtual d'imatges de teixits biopsiats amb aprenentatge profund


In extreme life or death situations, a quick and accurate diagnosis is essential to help pathologists look for signs of disease. Last August, engineers at the University of California (UCLA) unveiled a method to improve diagnostic tools that examine biopsied tissue samples. Through an artificial intelligence system, tissue images are a are virtual stained in a much faster and more accurate manner than previously.

Haematoxylin and eosin (H&E) staining is one of the most widely used in medical diagnosis. Under the microscope, pathologists examine tissue samples from biopsies stained with special dyes to improve colour and contrast. However, additional stains are sometimes needed to increase the contrast and colour of the components of the tissue. Although special stains provide a better diagnostic image, they often require much longer tissue preparations, higher control, and are more costly.

To speed up the process, UCLA researchers have created a computational technique with artificial intelligence that transforms images of previously stained H&E tissues into new ones with special stains added.

Less than a minute

According to the researchers, this process can be achieved with AI in less than a minute per tissue sample, compared to the hours currently spent. “Speed and accuracy are important in diagnosing medical conditions, such as organ transplant rejection. Rapid diagnosis enables rapid treatment”, said lead researcher Aydogan Ozcan during the presentation. “We have developed a technique based on deep learning that eliminates the need for special stains by histopathologists,” Ozcan added.

The team of scientists demonstrated the AI-based technique by generating a complete panel of special stains for kidney tissue, using deep neural networks trained in H&E stained tissue biopsy images. Since the technique is applied to existing images, the researchers stressed that it would be easy to adopt, as it does not require any change in the current tissue processing workflow used in pathology laboratories.