Natural language processing (NLP) is a branch of artificial intelligence (AI) to program computers to analyze human language via text. It is commonly used for translation services and question & answering systems such as Apple’s Siri or Amazon’s Alexa devices.
However, one of the great challenges for AI researchers is how to create models that are capable of answering questions that go beyond the text themselves, by applying additional common sense knowledge.
Ping An Smart Healthcare (PASH), in collaboration with Shanghai Jiao Tong University (SJTU), has developed an NLP solution for use by medical doctors that recently took first place in an international competition to test its capability for common sense inference. It is part of PASH’s AI-based medical decision support tool, AskBob, launched in April 2019.
AskBob is a ground-breaking medical decision-support tool driven by artificial intelligence (AI) to provide critical, up-to-date medical information to general medical practitioners. AskBob supports doctors during consultation and treatment of more than 1,500 diseases.
The NLP solution, running alongside PASH’s own AI algorithm platform, Safran, won the Commonsense Inference in Natural Language Processing (COIN) 2019 text comprehension competition of EMNLP-IJCNLP Conference (2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint Conference on Natural Language Processing). The PASH-SJTU team beat competitors including Baidu, Microsoft, Washington University, and the Indian Institute of Technology. The team ranked first overall, closest to the human standard, after achieving the highest scores in tests of understanding everyday language and news articles.
AskBob is already being used by more than 90,000 medical practitioners in more than 8,000 hospitals. It is available on the Chinese Medical Association's physician service platform and has been introduced to Singapore through collaborations with SingHealth, one of Singapore’s largest public healthcare groups, and the National University Health System (NUHS).
At the point of patient care, such as a hospital, AskBob connects with the medical IT system and extracts patient information. The AI engine then prompts the medical practitioner to enter symptoms and recommends appropriate laboratory tests. Drawing on the medical information, AskBob then makes a diagnosis recommendation for the doctor to confirm or change. After the diagnosis is made, AskBob provides a personalized drug recommendation.
AskBob uses Ping An Group’s Smart Medical Knowledge Graph, which combines millions of medical concepts, medical correlations and evidence in an unprecedented knowledge base across the health care ecosystem. The graph connects the dots between data points, providing the structure and context for the AI to “think”.