GBC is cooperating with Danner Lab of the US to establish AI for cervical cancer cell diagnosis. A panoramic smear is cut into digital cell images, and a liquid-based thin smear can be cut into about 2,300 slices . Cell classification is done according to the Bethesda System, and machine learning is used in an automated cell recognition system. This result can reduce the work pressure of medical staff and improve the accuracy of diagnosis to achieve a win-win effect
Population aging has become a common problem worldwide, and the incidence of cancer and chronic disease is increasing year by year. Fast and accurate diagnostic tools are therefore in demand to improve the quality of medical care.
A malignant tumor is a mass of abnormal tissue that grows beyond normal tissue with uncontrolled cell proliferation; moreover, it continues to grow regardless of the nutritional status of its host. An important feature of malignant tumors is the invasion of adjacent tissues and metastasis to distant organs and tissues through lymphatic and blood diffusion. However, it may also be spread by human factors such as surgical instruments, though this is extremely rare. Generally, malignant tumors grow rapidly with expansion invasion and metastasis. When observed with the naked eye, demarcations are not obvious while edges and sections are irregular. Necrotic tissues are common, and tissue often has no normal structure and increased mitotic frequency. Microscopic observation shows that the nucleus is irregular, enlarged, and has obvious nucleoli. Malignant tumor cells have low adhesion and peel off easily, and they may metastasize to other sites along blood vessels and lymph nodes. Clinical application of this property led to the invention of cytology cancer cell diagnostics. This method is non-invasive and is diagnosed by cell staining. Simple, convenient, and cost-effective, such cancer cell diagnosis is very successful and the associated prevention and treatment of cervical cancer in women is quite effective.
A “pap smear artificial intelligence (AI) digital image-assisted screening system” currently developed domestically involves a combination of three major fields encompassing domestic information, optics, and biomedical technologies; and associated products fall under the digital pathology system industry. Digital pathology is applied to cytopathology and drug development in Europe, the United States, and Japan. Mainstream market products include full-slide pathology scanning equipment, pathology image analysis and processing software, and pathology cloud management systems.
Artificial intelligence (AI) is a field of computer science that is committed to solving common cognitive problems related to human intelligence such as learning, problem solving, and pattern recognition. No longer confined to imaginary robots in science fiction, AI now constitutes a real application of high-level computer science that can improve the quality of medical care and improve human health.
Given more data, AI becomes “smarter” and learns faster; meanwhile, more accurate data makes machine learning and deep learning solutions more complete. Therefore, it is very important to have a wealth of clinical diagnostic cases.
Cervical cancer is currently the most effective cancer to prevent and detect among women in Taiwan. Pap smears are also recognized worldwide as the most effective tool for screening precancerous lesions and early-stage cancers. According to statistics, the three-year pap smear screening rate of women in Taiwan is more than 50%, compared with 80% of women in developed Western countries. There is thus still room for improvement in promoting pap smear screening and the domestic prevention of cervical cancer. The proportion of positive results for cervical screening is approximately 1.27%, and about 50,000 women who have not received a pap smear have cervical lesions (including precancerous lesions) without knowing it.
At present, domestic physicians and medical examiners use the naked eye and manual microscopes for screening, and there is still a small number of artificial diagnostic errors. However, many advanced countries in Europe and the United States use thin-layer smears and AI to assist in the interpretation and diagnosis. Medical laws of the United States and Taiwan stipulate that a cytologist cannot screen more than 80 pap smears in a single day, and manual reading thus has its limitations. This is in addition to an inability to record and save complete slice and smear image information; hospitals need to have enough space to preserve benign specimens for more than ten years, and abnormal smears are permanently stored. In order to improve the quality of medical diagnosis and reduce the workload of medical personnel, the benefits and requirements of AI development will then be seen in the digitization of pathology smears and the use of auxiliary screening systems.
The thin-layer smear of GBC/Danner Lab assists in the establishment of AI for domestic cervical cancer cell diagnosis. Compared with traditional smears, thin-layer smears have a single layer of cells free of impurities, mucus, and inflammatory cells. It is more suitable for the cutting of panoramic smears into computer cell images, as a liquid-based thin-layer smear can be cut into about 2,300 slices. The images are classified according to the Bethesda system (American system), and the computer is trained to engage in deep learning as an image parameter for machine learning of the automated cell recognition system.
It is hoped that this achievement can reduce the work pressure of medical staff and improve the correct diagnosis rate in the near future, achieving a win-win effect.