Cover Story
Image Recognition & Deep Learning for Enhanced Medical Treatment

Cervical cancer is a real fear for women today. Cervical cancer is the third most common form of cancer among women in the world, secondary only to breast and colorectal cancer. One woman dies every 2 minutes from the disease. In 2015, about 280,000 women died of cervical cancer. China is one of the countries with the highest incidences and mortality rates of cervical cancer in the world. There are over 130,000 new cases each year. 

AI, Resolving Biopsy Analysis

Healthcare involves both prevention and treatment. For prevention, many people choose to get the HPV vaccine. At the diagnostic level, a biopsy is the gold standard of diagnosis, and is the most reliable in all kinds of tumor examination methods. If the biopsy is performed incorrectly, all subsequent treatment will deviate. 

It is difficult for a pathologist to accurately match the sample against hundreds of cases. They always rely on experience. At the same time, limited by factors such as limited capital and insufficient management models, the reserve of talent in Chinese pathology departments is insufficient. According to the relevant regulations of the Ministry of Health, one or two pathologists are required for every 100 beds, and about 100,000 pathologists are required in total. However, as of 2017, the number of registered pathologists in China was only one tenth this number, and the number of those able to perform a biopsy in the cervical cytology was insufficient. The sensitivity is only about 65% due to factors such as fatigue, skill level, and subjective judgment.

In recent years,  AI has become an assistant for doctors. With the aid of AI image recognition and deep learning, it can improve the efficiency of the biopsy, improve the sensitivity and specificity toward lesion cells, and assist in analyzing a pap smear.

HUAWEI CLOUD Assists KingMed in Efficient Screening

KingMed was established 16 years ago, and has 37 medical laboratories, covering 34 provinces and municipalities throughout the country. They perform more than 50 million tests each year.

In the process of rapid business development, KingMed also noticed the contradiction between the serious lack of pathological doctors and the rapid growth of inspection in China. The number of registered pathologists is only one third that of the United States, despite having a huge population of more than six times the United States. This is the biggest bottleneck in all of KingMed's services. "A large portion of specimens received annually in KingMed are screened for cervical cancer, and most of them require a large number of views by pathologists." Li Yinghua, KingMed CIO, said in an interview: He believes that assisted screening through artificial intelligence can greatly reduce the amount of repeated work performed by doctors. It can be seen as an effective remedy for many pain points, improving screening efficiency, improving screening quality, and effectively solving the shortage of cervical cancer screening personnel and insufficient production capacity.

Li Yinghua said that they had contacted other cloud service providers. After a detailed selection process, they chose to cooperate with HUAWEI CLOUD because in addition to the high efficiency and reliability of cloud services, Huawei has a wide range of mature application cases in various industries. 

First, the medical industry has high requirements on data security and reliability. Currently, KingMed has extremely rich medical inspection data. The requirements for data transmission, storage, and analysis can be fully addressed by HUAWEI CLOUD.

Second, the HUAWEI CLOUD AI has a strong R&D team. In the challenging field of medical artificial intelligence, the HUAWEI CLOUD AI technical expert team and the pathological expert team at KingMed can make a strong joint effort and reach new height in the industry.

Sensitivity Exceeds 99%

Li Yinghua said that because the project involves cross-disciplinary knowledge, both parties have discussed and consolidated the terms and standards in detail. In order to convert the expert experience of senior pathologists into knowledge that can be learned by artificial intelligence algorithms, the two parties use a semi-automatic annotation platform using pre-annotation to effectively "see" the pathologist work, and to store this as data. This digitalized experience is converted into quantitative data.

Through just half a year of cooperation, the current AI solution for cervical cancer screening has made significant progress. Based on the production system, for the two types of evaluation, the AI model has a sensitivity of 99% and specificity of 60% or higher. "AI really can reduce the workload of pathologists, and free their hands for more challenging work." 

Breakthroughs are achieved in a short period of time because HUAWEI CLOUD has the platform, algorithms, and data required to apply AI in the medical industry.

First, at the platform layer, HUAWEI CLOUD AI has abundant computing resources for the fast iteration of AI models, and develops semi-automatic annotation tools based on the iterative AI algorithm model, improving doctor annotation and review efficiency by more than 500%. This greatly alleviates the difficulty of annotation in the medical imaging field.

Second, at the data layer, precision annotation data is the base of the deep learning model. KingMed is a leading enterprise in Chinese pathology. It has abundant sample data, authoritative senior pathology experts, and a comprehensive base of precise annotation data.

Third, the artificial intelligence algorithms in HUAWEI CLOUD AI are continuously iterated upon and are leading in model precision and speed. Recently, the HUAWEI CLOUD AI medical team has invested heavily to resolve major technical problems in the medical field, and has won competitions in  many medical events, such as LUNA-2016, HC-2018, and ISLES-2018.

Cancer may be seen as an endless night for some, but AI is helping give us a glimmer of dawn. The HUAWEI CLOUD AI medical solution deeply integrates Huawei's artificial intelligence technology and medical expert experience. Through analysis of in-depth clinical scenarios and doctor work processes, we can solve the most critical issues faced by doctors, free them from repetitive work, improve testing efficiency, solve the extreme imbalance between doctors and patients, and improve testing accuracy.

In the future, the HUAWEI CLOUD AI medical team will continue to explore the application of artificial intelligence technologies in medical imaging fields such as ultrasound, pathology, X-ray, CT, MRI, and endoscopy. They will continuously explore possible applications of artificial intelligence in various medical scenarios. By integrating multi-modal information such as imaging groups, genomics, and clinical groups, HUAWEI CLOUD will develop comprehensive solutions based on end, edge, and cloud collaboration, such as auxiliary image diagnosis, chronic disease management, health management, drug analysis and mining solutions. The technical and social benefits of AI in the medical field will be continuously elaborated upon.

Provides Scalable, On-demand Computing Resources

Learn More