The Japanese Society of Pathology, the National Institute of Informatics, the University of Tokyo, and others have jointly developed a pathological diagnosis support AI that determines the presence or absence of a tumor (cancer) from the pathological tissue image of a stomach biopsy.
Microscopic pathological diagnosis by a pathologist is essential for confirming the diagnosis of cancer. Diagnosis) risk is regarded as a problem.
Therefore, our group has been working on the development of AI that double-checks pathological diagnoses, and this time, we have achieved a diagnosis match rate of 90 to 97% with pathologists, targeting gastric biopsies, which are the most frequently used specimens in routine pathological diagnoses. Succeeded in developing a pathological diagnosis AI that reaches
For AI image recognition machine learning, we used deep learning, which has excellent performance in image recognition.In addition, the images used for pathological diagnosis are full-color and ultra-high resolution, and the amount of information is an order of magnitude greater than that of general image recognition. ” was developed and adopted.
This method mimics the way a pathologist makes a pathological diagnosis by combining low and high magnification observations under a microscope. Machine learning can be done without losing global location information.According to the company, this method can reduce false positives and can determine cancer and non-cancer with higher accuracy than the conventional method.
By using this AI to support double-checking in the field of pathological diagnosis, it is expected to reduce the burden on pathologists and improve diagnosis.In addition, by incorporating it into the remote pathological diagnosis network, it is possible to promote the equalization of cancer medical care (equal access to advanced medical care anywhere in the country).
Paper information:[Cancer Science] Development and multi-institutional validation of an artificial intelligence-based diagnostic system for gastric biopsy
