アクセス　Laboratory of Histopathology and Cytopathology Gunma University Graduate School of Health Sciences

About pathology and AI
What do you think about artificial intelligence?

I will try to explain it by answering some frequently asked questions.

Q1: As AI develops, won't the work of cytotechnologists become unnecessary?

A: No. Digital image information includes information such as the three primary colors (red, blue, and green) and the vertical and horizontal position information of each color tone on every dot (called a pixel) on the image.It stores 5-dimensional information as numerical information.Learning by artificial intelligence involves converting the high-dimensional information contained in each pixel of the image by repeating it about 20 to 50 times using various mathematical formulas. This is called folding.
This folding compresses high-dimensional information into two dimensions that are easier for humans to handle. What is important is that ``no medical information, such as the structure and function of the human body, cell characteristics, or pathological basis for diagnosis, is used in these processes.'' Simply mathematically simplified information such as ``This is a case that metastasized, this is a case that did not metastasize'' is taught (this is called learning), and ``This is a case that metastasized, this is a case that did not metastasize.'' What are its characteristics?'' When this learning is done, if you tell the robot that this image is stomach cancer, this image is colon cancer, etc., the AI can also infer the disease name of an unknown specimen based on what it has learned. However, it is important to note that ``AI does not use any medical basis for image learning; it estimates unknown information based on numerical information that has been simplified through simple mathematical folding. ``Even if the unknown sample is something other than the learned content, it attempts to produce results within the range of the learned content.'' In other words, AI makes inferences from the data it knows, as if it knew everything. And he's full of confidence. ``In other words, it is possible to be confident but still make mistakes.'' how is it? Knowing this, do you want to have AI diagnose your pathological specimens? I absolutely hate it. A statement from the American College of Pathologists declares that the use of AI for pathological diagnosis is not possible.

Q2: So AI is not used in the pathology field?

A: That's not true. For extremely simple judgments such as the depth of a color tone, you can learn that this is no stain at all, this is light stain, this is intermediate stain, and this is dark stain, and then you can tell which of these is the stain of an unknown specimen. Determining whether something is applicable is a special skill of AI. People may make different judgments about the same specimen when they are sleepy, tired, or alert, and each person perceives the same image differently, so It may actually be better to leave simple tasks to AI. Therefore, in the future, it will become possible to quantify information such as simple color tone evaluation and color area evaluation in the field of pathology, and it is expected that the amount of information disseminated from the field of pathology will increase.

About the future of Medical technologist in the field of pathology

Q1: As automation progresses in the field of pathology, will the work of pathology Medical technologist be replaced?

A: Please feel free to proceed to the path of becoming a Medical technologist in the pathology department. The specimens handled in pathology are the cancer focus or the part of the disease that is in the midst of the disease (this is called the lesion). For example, cancer is composed of not only cancer cells but also a wide variety of cells such as lymphocytes, histiocytes, fibroblasts, vascular endothelium, and smooth muscle cells. Even if you perform a genetic search on this lesion, you will only get data that contains a mixture of data on cells other than cancer cells. To avoid this, it is necessary to treat the tissue with enzymes and purify it to only cancer cells. Yes, it is very time-consuming. Not every hospital can do it. On the other hand, with pathology specimens, only cells that are morphologically known to be cancerous can be examined, and samples from any hospital in the country can be analyzed. Since genetic testing is expected to be performed by limited institutions and in limited cases, it is expected that pathological methods will continue to be used frequently.

Now, as mentioned above, pathology will become increasingly digitalized in the future. If this happens, the analysis of the cancer itself will progress rapidly and we will be able to provide a large amount of information. In addition, pathological specimens are diverse. Uniform mechanization is extremely difficult. Since the process involves a wide variety of specimens, from about 1 mm to tens of centimeters, it cannot be easily mechanized.

In other words, it is conceivable that the work that I have been doing as a clinical laboratory technician will remain, while the number of specimen preparations for a wide variety of analyses will increase. Even within the field of clinical testing, the role of the pathology department as the department that deals with the ``essence of the disease'' will never waver.

   (Written by: Masanao Saio)