IBM Watson Health Files IPO Prospectus: Pioneering AI-Driven Clinical Decision Support Since 2007

Who is Watson?
Watson is an artificial intelligence system developed by the DeepQA project team led by IBM’s chief researcher, David Ferrucci, since 2007. It represents the culmination of four years of dedicated effort by more than 20 IBM researchers and is named after IBM’s founder, Thomas J. Watson. Thanks to the scientists’ endeavors, Watson possesses the ability to understand natural language and provide precise answers to questions.

In 2011, Watson defeated top prize winner Brad Rutter and winning-streak record holder Ken Jennings on the long-running American quiz show *Jeopardy!*, thereby entering the public spotlight.

According to IBM, Watson’s hardware consists of a cluster of 90 IBM Power 750 servers, comprising a total of 2,880 Power7 processors and 16 TB of memory. On the software side, Watson is written in Java and C++, and leverages the Apache Hadoop framework for distributed computing, along with the Apache UIMA (Unstructured Information Management Architecture) framework, IBM DeepQA software, and the SUSE Linux Enterprise Server 11 operating system.

Why Is Watson So Smart?
How Did Watson Manage to Defeat Humans in Competition? How Did It Become So Intelligent? Watson stored millions of documents, including dictionaries, encyclopedias, news articles, literary works, and other reference materials that could be used to build a knowledge base. Watson’s hardware configuration enabled it to process 500 GB of data per second, equivalent to reading one million books in a single second.

According to Dr. Zhang Lei of IBM China Research, upon receiving a query, Watson performs a series of computations, including syntactic and semantic analysis, searching across various knowledge bases, extracting candidate answers, seeking evidence for these candidates, and calculating and synthesizing the strength of the evidence. The core technical principle of Watson involves searching numerous knowledge sources and applying a multitude of small-scale algorithms from multiple perspectives to comprehensively evaluate and learn from various possible answers. A critical step in this process is assessing the reliability of candidate answers by detecting and quantitatively evaluating their credibility across multiple dimensions, such as keywords, geographic location, and entity type.

The Journey of Watson?
In 2011, IBM stated that, given Watson’s ability to understand human language, it could rapidly provide diagnostic hints and treatment recommendations by inquiring about patients’ symptoms and medical history, leveraging artificial intelligence, natural language processing, and analysis technologies, and drawing on information and data collected from various sources. Subsequently, an Associated Press reporter experienced firsthand how the robotic doctor Watson diagnosed patients. Developers fed Watson case details of a virtual patient with eye conditions one by one: blurred vision, a family history of arthritis, residence in Connecticut, pregnancy, and more. Watson proposed several differential diagnoses, including uveitis, Behçet’s disease, and Lyme arthritis. It was estimated that Watson’s diagnostic accuracy rate reached 73%.

In 2011, U.S. healthcare services provider Wellpoint signed an agreement with IBM, marking Watson’s first professional engagement. Watson’s primary tasks were to assist Wellpoint nurses managing complex cases and to review medical requests from healthcare providers, with subsequent applications in oncology clinical trials. The President of Wellpoint also noted that, in the future, Watson might be able to access patients’ medical records and other relevant information, synthesizing and feeding this data back to physicians to accelerate diagnostic processes.

In 2014, The University of Texas MD Anderson Cancer Center partnered with IBM to launch the “Moon Shot” initiative, which leverages IBM Watson technology to combat cancer. As the top-ranked and globally recognized leading cancer hospital in the United States, MD Anderson Cancer Center developed the Oncology Expert Advisor (OEA), a system powered by the Watson cognitive computing platform, designed to integrate the knowledge of the center’s clinicians and researchers. The OEA system assists clinicians in formulating, monitoring, and adjusting treatment plans for cancer patients. IBM Watson technology also facilitates the simplification and standardization of the collection and integration of patient medical records, laboratory data, and research data, enabling these consolidated datasets to be incorporated into MD Anderson’s centralized patient database for in-depth analysis using advanced analytical techniques.

In addition to the Anderson Cancer Center, the Mayo Clinic is also conducting concept trials with IBM Watson to provide patients with suitable clinical trials more quickly and efficiently. At any given time, the Mayo Clinic can conduct over 8,000 human research trials. However, many clinical trials fail to be completed due to insufficient participant enrollment, a challenge faced both at the Mayo Clinic and elsewhere. IBM and the Mayo Clinic are expanding Watson’s knowledge corpus by incorporating data from the Mayo Clinic and public databases such as ClinicalTrials.gov, while training the system to analyze patient records and clinical trial criteria to provide appropriate matches.

Bumrungrad International Hospital in Thailand is leveraging IBM Watson cognitive computing at its Bangkok research center to enhance the quality of cancer care and conduct case evaluations across its facilities in 16 countries. The hospital has committed to utilizing Watson for Oncology, a technology co-developed with Memorial Sloan Kettering Cancer Center (MSK), over the next five years. This system will assist physicians in formulating effective treatment plans for cancer patients by integrating medical evidence, academic research, MSK’s extensive clinical expertise, and individual patient records.

In addition, Memorial Sloan Kettering Cancer Center in New York and the Cleveland Clinic also have business collaborations with Watson.

In the corporate sector, Johnson & Johnson leverages IBM Watson to read and comprehend scientific papers detailing clinical trial results, using these insights to formulate and evaluate drug treatment regimens and other therapeutic approaches. With this capability, Watson Discovery Advisor helps scientists identify whether any adverse genetic profiles are present in drug samples. In contrast to previous comparative studies, which required three individuals an average of ten months to collect and prepare data before analysis could begin, the Johnson & Johnson team aims to rapidly synthesize useful information directly from medical literature using Watson, enabling them to pose queries directly to the data.

According to Gamp, a member of the VCBeat startup community, IBM US underwent a restructuring this month, dividing the company into seven major divisions, one of which is the healthcare division centered around Watson.

The Development of Artificial Intelligence in the Healthcare Industry
In fact, long before Watson, healthcare institutions were already developing projects for “computer doctors.”

In the 1970s, researchers at the University of Pittsburgh in the United States developed a software program called “Quick Medical Reference” for diagnosing complex conditions in general internal medicine. This expert medical diagnostic system compiled 4,300 clinical manifestations and was capable of diagnosing more than 600 diseases, thereby enhancing the likelihood of rapid diagnosis through systematic computational processing.

In 1972, Stanford University in the United States began developing the MYCIN system, which was basically completed and put into application in 1974. As “-mycin” is a common suffix in the names of many antibiotics, MYCIN is an expert system designed to assist physicians in diagnosing hospitalized patients with bloodstream infections and selecting appropriate antimicrobial therapies. It remains highly representative to this day.

In the 1980s, Massachusetts General Hospital in the United States began developing and refining the DxPlan project. The knowledge domain covered by DxPlan included most diseases and clinical manifestations across various internal medicine specialties, with IBM personal computers serving as the primary platform for program development. Users could consult the system to determine the next appropriate tests and examinations, thereby obtaining the maximum amount of information at minimal cost.

In the fall of 2010, the “Isabel Healthcare System” was integrated into the network at Orlando Health in Florida, USA, providing physicians with reliable diagnostic and treatment recommendations. Less experienced clinicians with limited practical exposure derived greater benefit from the system. However, Isabel could only be accessed through integration with multifunctional hospital healthcare systems, resulting in slower performance and relatively high costs.

By 2013, IBM Watson had entered the medical arena, serving both as an expert in cancer diagnosis and as a specialist in healthcare utilization management. Since then, computer-aided diagnosis has turned a new page, officially ushering in the “Watson Era.”

According to data from market research firm Insight Research, the U.S. healthcare industry will invest $69 billion in information technology over the next six years. Reports indicate that Intel and SAP have already begun collaborating with researchers at the University of California, Berkeley, to develop competitive medical supercomputers.

Are AI Doctors Really Reliable?
As a computerized physician, Watson possesses inherent advantages: its knowledge base far exceeds that of humans and is never forgotten; it maintains high diagnostic accuracy and can provide consultations continuously without rest. Although this appears encouraging, it has inevitably sparked various doubts. Abraham Verghese, a writer for The New York Times and a physician at Stanford University, stated, “Watson may potentially become an intelligent companion in our midst, but what I hear from patients, relatives, and friends is not a lack of technology, but rather an excess of it.” Marty Kohn of the Watson project team believes that Watson is merely an auxiliary tool; if physicians are unwilling to change, Watson alone cannot transform the healthcare industry. He remarked, “Some technologies have indeed changed medicine by providing treatments that were previously unavailable, but information technology (IT) is not such a technology. I believe IT is merely an enabler.” Klaus-Peter Adlassnig, a computer scientist at the Medical University of Vienna and Editor-in-Chief of the journal Artificial Intelligence in Medicine, argues that computerized physicians like Watson are, in essence, search engines capable of answering questions posed in natural language. Over time, although computers can learn from their errors, the knowledge they derive from medical literature and case studies remains shallow and broad. Consequently, such knowledge may not necessarily hold significant value for healthcare professionals in clinical settings.

The practice of medicine is far more complex than merely processing data. Whether it involves providing emotional support to patients and their families, navigating subtle nuances in clinical practice, or learning to manage uncertainty, every aspect relies on human physicians. Even the most extensive medical datasets and the most advanced computational capabilities cannot teach a computer how to care for patients with the same compassion as a human doctor.

Watson is not omnipotent. Although it has gained widespread renown, it ultimately serves only as a physician’s assistant; the vision of an “AI doctor” still has a long way to go. Perhaps in the future, human–machine collaboration will be the ultimate direction for development.

(To stay updated on the latest internet healthcare startup news, please follow VCBeat’s WeChat official account: vcbeat. We also welcome you to engage with us on topics of interest, contact us via WeChat, and share your startup projects or related research insights.)