Healthcare

Driving the usefulness of AI in healthcare: IBM Research showcases work at AMIA 2020

Share this post:

Medical Informatics is the science behind how to use data, information and knowledge to improve human health and the delivery of health care services. It is a growing discipline that is seen as the key to accelerating the goals of healthcare reform and the promise of personalized and precision medicine. AMIA (American Medical Informatics Association), as both an organization and a scientific community, is committed to the vision of a world where informatics transforms health care. AMIA is also the professional home for the informaticians of today and the driver of informatics’ future. IBM is deeply engaged and committed to this endeavor and to the upcoming AMIA Annual Symposium which will be held virtually November 14 – 18 .

Researchers from our IBM Research labs around the world and from IBM Watson Health have contributed a total of 47 workshops, papers, posters and panels that will be presented at the upcoming meeting. These contributions cover a wide range of topics but reflect our overarching goal of driving the usefulness of AI in healthcare. Here are some highlights:

Improving the Fairness and Responsibility of AI

At a time when AI is poised to make major contributions to healthcare, there is an increasingly urgent need to develop responsible AI. One aspect of responsible AI Is leveraging AI methods to help vulnerable populations and improve health equity. In “Identifying Factors Associated with Neonatal Mortality in Sub-Saharan Africa using Machine Learning” IBM Research reports on an important study to identify factors associated with neonatal mortality. As part of a larger collaboration with the Bill and Melinda Gates Foundation, the IBM authors of the paper, who are based in Nairobi, Kenya, analyzed complex and broad data from 10 sub-Saharan countries, including two recent Demographic and Health Surveys (DHS) on women who recently suffered a neonatal death; the MAL-ED dataset looking at connections between malnutrition, gut and growth; as well as the AMANHI study into direct factors of neonatal death.

This work led to uncovering data that could lead to understanding why early childhood deaths occur. Researchers built algorithms to break out of traditional and often unexplainable models of AI analysis, as well as dig into variable social factors that extend beyond the clinic. By untangling the complex heterogenous makeup of Africa’s maternal population, researchers used AI to find direct correlations between birth spacing (the number of years between pregnancies) and the size of a woman’s social support network on birth outcomes.

In another effort reported in “Visualizing Inequities in Clinical Trials using ML Fairness Metrics”, IBM Research collaborated with researchers from Rensselaer Polytechnic Institute in a study funded by the IBM AI Horizons Network to develop a Machine Learning (ML) method to examine the representation of protected classes of subjects in Randomized Clinical Trials, and to provide guidance in designing more equitable clinical trials.

Another increasingly urgent aspect of responsible AI is the evaluation and assessment of ML methods in the context of clinical use, particularly to avoid unintended harm. For example, one type of machine learning method called Deep Reinforcement Learning (DRL) has attracted much attention, partly because this is the method behind the success of AlphaGo. It has been applied to medical domains such as decision support in the ICU. However, the quality and reliability of such models in clinical settings has not yet been fully assessed. To help do so, IBM researchers collaborated with researchers from MIT and Harvard to carry out a study funded by the MIT-IBM Watson AI Lab to systematically explore the sensitivity of DRL for ICU use cases, and discovered important areas of caution. To emphasize the considerations that should be kept in mind to prevent this from happening, this and other findings will be presented in the paper “Is Deep Reinforcement Learning Ready for Practical Applications in Healthcare? A Sensitivity Analysis of Duel-DDQN for Hemodynamic Management in Sepsis Patients”.

Progressing the Understanding of Type 1 Diabetes with AI

One way for AI to be used to answer meaningful questions in healthcare is for ML researchers to work closely with clinical domain experts. IBM Research has a long tradition of forming partnerships with research foundations such as The Michael J. Fox Foundation and CHDI for this purpose.

In 2017, IBM Research started a collaboration with JDRF to research risk factors for Type 1 Diabetes (T1D) in children. Through this collaboration, IBM Research played an instrumental role in the formation of a new international research consortium, The T1DI (Type 1 Data Intelligence) Study. The T1DI Study represents a collaboration spearheaded by JDRF and IBM Research, with participation from eight academic centers in four countries. As part of IBM Research’s larger mission to launch and support communities of scientific discovery, this initiative is aimed to help accelerate the speed at which this serious condition can be understood.

The IBM Research team aggregated data from five different disease registries collected from longitudinal studies conducted over the past thirty years in four countries, and led the development of multiple machine learning models to study the heterogeneous evolution of Islet autoantibodies. This has led to a better understanding of underlying etiology of T1D, which could potentially help to develop better treatment screening. The collaboration has already led to a series of new findings presented, at ADA2019, ADA2020, EASD2020 and ISD2020. Three new research reports will be presented by IBM Researchers at AMIA2020, which further progress the scientific community’s understanding and work towards better prediction of the disease’s onset. These papers include:

Improving AI’s Use in Healthcare 

The ability to learn directly from data is one of the key attributes of AI that led to its much-publicized successes in recent years. However, there is now increasing realization that data is almost always limited: in quantity, quality, coverage and biasedness among other measures. In the healthcare domain in particular, the injection of clinical knowledge is critical to ensure the usefulness of what is learned from data with full awareness of its limitations, and one way this is done in AI is through knowledge representation and reasoning.

Many IBM Research presentations at this year’s AMIA symposium reflect our significant efforts and progress made on this front, including “Accelerating Epidemiologic Investigation Analysis Using NLP and Knowledge Reasoning: A Case Study on COVID-19”, “Knowledge Extraction and Prediction from Behavior Science Randomized Controlled Trials: A Case Study in Smoking Cessation” , “Combining User Preference and Health Needs in Personalized Food Recommendation” and “Combining Deep Learning and Knowledge-driven Reasoning for Chest X-Ray Findings Detection”.

Working Towards Adaptive Healthcare Systems with AI

Finally, COVID-19 has sent shock waves through the healthcare system and highlighted the many urgent needs where AI can play an important role, from accelerating clinical trials, to scaling up deployment of telemedicine, to opening up science strategies for developing therapeutics.

These and other topics will be discussed by a prominent panel of specialists at the Industry Sponsored Panel “COVID-19 – The Challenge, the Response, and Insights into the Future of Healthcare” hosted by IBM Research and IBM Watson Health.

In addition, the strains on healthcare caused by COVID-19 reaffirmed the need for an adaptive healthcare system that extends far beyond the four walls of the hospital, not just in response to a crisis, but as a well-planned long-term strategy to enhance the reach and quality of care. Two presentations from IBM Research highlight our efforts contributing to this strategy. “Artificial Intelligence Decision Support for Medical Triage”, which has been piloted by a leading telemedicine leader since 2019, and “Towards motor evaluation of Parkinson’s disease patients using wearable inertial sensors”.

Inventing What’s Next.

Stay up to date with the latest announcements, research, and events from IBM Research through our newsletter.

 

IBM Fellow and Global Science Leader, AI for Healthcare, IBM Research

More Healthcare stories

Getting AI to Reason: Using Neuro-Symbolic AI for Knowledge-Based Question Answering

Building on the foundations of deep learning and symbolic AI, we have developed a software able to answer complex questions with minimal domain-specific training. Initial results are encouraging – the system achieves state-of-the-art accuracy on two datasets with no need for specialized training.

Continue reading

IBM Research at EMNLP 2020

At the annual Conference on Empirical Methods in Natural Language Processing (EMNLP), IBM Research AI is presenting 30 papers in the main conference and 12 findings that together aim to advance the field of natural language processing (NLP).

Continue reading

DualTKB: A Dual Learning Bridge between Text and Knowledge Base

Capturing and structuring common knowledge from the real world to make it available to computer systems is one of the foundational principles of IBM Research. The real-world information is often naturally organized as graphs (e.g., world wide web, social networks) where knowledge is represented not only by the data content of each node, but also […]

Continue reading