Google Develops Augmented Reality Microscope to Better Detect Cancer

Google Develops Augmented Reality Microscope to Better Detect Cancer

Google AI researchers developed an artificial intelligence (AI) system that can accurately detect cancer — prostate and breast cancer so far — in a standard microscope in real time.

Besides its potential with cancer, this technology may be helpful in diagnosing other diseases and in several other research areas.

Findings of the study, titled “An Augmented Reality Microscope for Real­time Automated Detection of Cancer,” were shared in the Google AI blog and recently presented at the Annual Meeting of the American Association for Cancer Research.

Microscopic review of tissues, such as biopsy samples, on glass slides by a pathologist — a physician specializing in disease diagnosis based on examination of body tissues and fluids — is the current gold standard for cancer diagnosis.

However, cancer diagnosis is a time-consuming process because it can be like searching for a needle in a haystack, involving the screening of a vast amount of data, and tumors often resemble healthy tissue.

Digital pathology transforms tissue samples on glass slides into digital images, which have been used to create AI algorithms that improve diagnostic accuracy. However, digital pathology systems are still very expensive, limiting the widespread adoption of AI algorithms in pathology.

Now researchers at Google AI have developed an approach called the Augmented Reality Microscope (ARM) that brings AI algorithms to the regular microscope.

It consists of a standard light microscope with a slight modification — the addition of a small digital camera and augmented reality display connected to a computer that runs the algorithm — which can be performed in any microscope in the world using low-cost, readily available components.

This allows the real-time visualization and analysis of an image in the microscope by the AI algorithm at the same time as the pathologist. Then, the results of the AI algorithm are directly shown in the field of view/image by highlighting the areas that look like cancer, so the doctor can pay special attention to them.

To test this technology, the researchers configured it to run two different cancer detection algorithms: one that detects prostate cancer in prostate samples, and another that detects breast cancer metastases in lymph node samples. These algorithms were developed through the analysis of thousands of digitally scanned images, where they were “trained” to recognize cancer cells.

Both AI algorithms showed great accuracy (above 95%) in detecting cancer cells in the ARM. Researchers believe the algorithms’ performance may be further improved if they are “trained” directly on microscope images, instead of digital ones, as they have different optical configurations.

When evaluating the speed and accuracy of cancer diagnosis of six U.S. board-certified pathologists with and without the breast cancer AI algorithm, the researchers found that the AI algorithm was able to improve both parameters.

“The main potential of the ARM is in expanding access to AI to users who are not likely to adopt a digital workflow in the near future,” the team said.

The ARM can be potentially useful for research and clinical usage, such as quantification of immune cells, immunotherapy biomarkers, and dividing cells, measurement of tumor size, and identification of bacteria and infections.

Researchers noted that a larger study is required to evaluate the algorithms’ performance with higher statistical significance, as well as a diagnostic accuracy study to assess the technology’s clinical performance by a group of pathologists.

In the future, they also imagine a system with robotic components outside the microscope that would allow automated exhaustive “scanning” of the tissue samples to detect tumor or metastasis.