Developing Concept Enriched Models for Big Data Processing Within the Medical Domain

Introduction

Though the human brain is an incredibly complex system, it has its own limitations on the amount of information it can synthesize and recall. However, through the aid of cognitive computing, human cognition can be supplemented with computer systems that implement semantic and neural models of human thought (Wang et al., 2018). Through the application of cognitive computing to a plethora of diverse domains, new advances once unfathomable can hopefully be attained. For example, in 2018 the first reports emerged of artificial intelligence performing better than humans on a medical clinical examination. On June 28th, 2018, Dr. Mobasher Butt stood on stage in London’s Royal College of Physicians, where he announced that his company’s trained AI received a score of 82%, beating out the average by medical students of 72% (Olson 2018). Dr. Ali Parsa, the founder of Babylon Health, states that on the planet, over 5 billion people lack the access to basic surgery. He claims that the United States has shifted its focus from health care to its economic benefits, and that there are large gaps in the health-care system. To fill these gaps in its healthcare infrastructure and services, Parsa predicts that the United States will be the largest consumer of artificial intelligence in healthcare in the near future (Olson, 2018).

While many other domains already use artificial intelligence to enhance the quality of life for their users, medicine has yet to make the breakthrough for various reasons. Cognitive Computing for the medical field heralds an era of change rapidly approaching. Starting in 2016, the U.S Department of Veterans Affairs (VA) hospitals in Durham, North Carolina, have used IBM Watson to help with diagnosing cancer patients by collecting DNA from tumors and analyzing the genetic material to determine possible causes as well as effective treatments. The VA treats nearly 4% of U.S. cancer patients, allowing IBM Watson to have a large sample size (Moscaritolo, 2018). Dr. Kyu Rhee claims that ”it is incredibly challenging to read, understand, and stay up-to-date with the breadth and depth of medical literature and link them to relevant mutations for personalized cancer treatments”; this sentiment is shared by many medical professionals, justifying the need for effective usage of artificial intelligence in the crucial domain of medicine (Moscaritolo, 2018).

In this paper, we examine existing cognitive computing technologies and the process for developing models to optimize them specifically for practical use in medical environments. While limited technologies currently exist for every-day clinical usage, the field remains wide open and a large, untapped market exists for new technologies to emerge (Gudivada and Tabrizi, 2018). In an unprecedented era where data is abundant yet largely under-utilized, the time to make advancements has arrived (Ahmed, Toor, O’Neil, & Friedland, 2017).

The rest of the paper is organized as follows:

• •

  Motivation

• •

  Related Work

• •

  Data Processing Tools

• •

  Building a Custom Concept-Enriched Model

• •

  Model Enrichments

• •

  Results

• •

  Future Work

• •

  Conclusion

This paper reflects a partially revised and updated version of the authors’ research originally published at the 18th IEEE International Conference on Cognitive Informatics & Cognitive Computing (2019).