Abstract assisted by AI cervical screening model interpretation 10 times faster than manual interpretation of
[China, Guangzhou, June 11]
Golden clouds Medicine and Huawei jointly announced in 2019 China biological industry conference, the two sides achieved a breakthrough in artificial intelligence (AI) aid pathological diagnosis application development. Golden medical pathology expert team with Huawei cloud AI team, for the first time based on the pathological, deep learning through technology, training a precise and efficient AI assisted cervical cancer screening model.
The model in the exhaust of the female is higher than 60%, based on the negative plate is the correct interpretation higher than 99%, while the detection rate of more than 99.9% positive disease. This is the highest level of AI assistance cervical cancer screening current international published. Typically, the cell pathologist read microscopic cervical smears, per patient takes 6 minutes, whereas AI identified only 36 seconds. AI cervical screening through the auxiliary speed manual interpretation model interpretation 10 times.
AI auxiliary stage cervical cancer screening to take a major breakthrough
Pathological diagnosis by the medical profession recognized as a disease diagnostic "gold standard." At present, cervical cancer screening for women age population over 350 million people, even at 3 - to conduct a screening to measure the standard five years, the level of screening is still far failed to meet the requirements. In addition to the popularity of the concept has been lacking, and more important reason is that the pathologist cells scarce, the ability to handle samples bottleneck, thereby constraining further promote cervical cancer screening.
2018, Golden Medicine and Huawei signed a cooperation agreement, cooperation in the field of artificial intelligence. The Golden Mile with Huawei cloud collaboration focused on the development of cervical cancer screening model of AI assistance. According to Dr. Luo Pifu director of the Center for Medical Pathology Golden Mile reports, this development based on the calendar year 43.5 million cases of cervical cytology screening sample Golden Mile, pick out the nearly 200,000 tiles, for accurate labeling and AI assisted screening model training. The validation set has been used more than 20,000 cases of sample data, the model is validated AI.
Verification results show that the model of the female rows 61.9% reduction in interpretation doctors doing work; negative sheet interpreted correctly higher than 99%; positive detection rate of 99.9% over the lesion.
The powerful combination of hard power to support AI model development
取得如此重大的突破，金域医学和华为云的专业知识、样本数据、AI技术与算力等要素不可或缺。
金域医学作为全国领先的第三方医学检验机构，截至2018年，金域宫颈病变检测中心累计收到来自全国31个省市区的宫颈癌筛查样本超过4350万例。“此次采用的训练集和验证集，规模是行业平均水平的数倍以上，极大地保障了结果的可靠率。”罗丕福博士说。
此外，金域的病理医生优势也发挥了关键作用。目前，金域有全职病理医生近500名，其中从事细胞病理学的医生近 300名。此次，金域投入了多位病理专家全职负责审阅与标注工作，以2人一组的模式进行互相 验证，相当于由病理专家 “手把手” 向AI模型传授病理诊断知识。
在AI技术支持方面，华为云一站式AI开发平台ModelArts，从标注、训练模型、图片判断、推理等方面为AI模型的训练和应用提供了坚实的基础。
在标注阶段，借助华为云ModelArts对已经上传到云端的海量图像块进行半自动化标注，大大节省了病理医生标注的时间和精力。
在算法层面，华为云不断迭代起来的人工智能算法在模型精度和速度上都达到了业界领先的水平。
近期，华为云AI 医疗团队投入重兵聚焦解决医疗领域的重大技术难题，且已经在肺结节检测比赛LUNA-2016、超声测量比赛HC-2018、脑中风分割比赛ISLES-2018 等多个医疗领域的顶尖赛事上获得业界领先水平。
华为云EI服务产品部总经理贾永利表示，深度学习技术的发展，使得AI辅助病理识别成为可能，而华为云强大的算法、算力和一站式AI开发平台，结合金域的大规模样本和病理学专家丰富经验，最终使AI达到了病理专家‘助手’水平。此次重大突破，正是AI与专家经验场景有效结合的证明。“未来，华为云AI将努力让AI服务触及每一位开发者、企业，共同推进行业智能化升级。”
效率大幅提升 未来AI辅助筛查应用可期
If the results of the application to the floor, is expected to enable the pathologist to make a single cytology workload reduced by more than 60%, significantly increasing screening efficiency. Typically, the cell pathologist read microscopic cervical smears, per patient takes 6 minutes, and the current AI identified only 36 seconds in the future there is further room for improvement.
At the same time, prompted by the AI and results than on reflection, the doctor can be more convenient to make a more accurate interpretation, quality control and training in pathology students cervical cytology can bring effective help. Once the AI assist future cervical cancer screening be promoted, people will be able to greatly enhance the range of services covered by the frequency of cervical cancer screening services, making the quality of screening women age population may be approaching the level of developed countries, to promote early screening of cervical cancer early treatment.
Golden Mile Medical Chief Information Officer Li Yinghua said, then, Golden cloud and Huawei will continue to invest more sample data, AI model for large-scale validation, at the same time more professional labeling to improve the model, while ensuring sensitivity premise under it specifically has been increasing. "We will also improve the reliability of the model so that it can adapt to the reagents and supplies a wider range of production levels, will promote the application of AI secondary screening to primary health care closer to the level of the scene."
It is reported that this outcome also laid a solid foundation for AI in the medical field in a wider range of application scenarios. In addition to mass spectrometry analysis, gene sequencing, molecular pathology, medicine and Huawei cloud Golden Mile will also seek new directions of cooperation in other pathological AI scenarios, such as gastrointestinal disease, kidney disease, blood disorders, rare diseases and other subspecialties scene to explore more applications possible.
Source: Huawei cloud community Author: Huawei cloud headlines