Written By: Kelly Buckman
Most of us in the healthcare IT field are familiar with the concept of Artificial Intelligence, or AI. AI attempts to harness the ability of computers and machines learn and perform tasks in order to automate and increase the efficiency of business processes. AI can be grouped into three separate areas of study: deep learning, machine intelligence, and machine learning. Here I will focus on the last of these, machine learning.
Machine Learning automates the analysis of data and inputs to develop computing models needed to operate an automated computer, software program or process, thus eliminating the need to do this manually, saving time, energy, and resources. Sounds great, right? It’s easy to embrace this technology as the way of the future, a means of achieving more efficient operations. However, it’s important to understand the limitations and risks associated with this technology.
-Machine Learning is susceptible to errors and when errors do occur, it can be more difficult to diagnose and correct them.
-Time and accurate data sets are needed to make predictions using Machine Learning, as “learning” is achieved through historical data.
-For scenarios that are not included in the historical data, it is difficult to prove beyond doubt that the predictions made by the Machine Learning system are applicable.
-Machine Learning cannot always provide rational reasons for a particular prediction or decision, as an incomplete or poorly written algorithms do not understand context and are prone to hidden and unintentional biases, depending on the data they have to work with. Furthermore, since Machine Learning is intended to answer questions and not ask them, human collaboration is important in order to better evaluate the output.
This brings us to the question of how Machine Learning can best be applied to Healthcare, a field where mistakes, poor judgement, and inefficient use of time can have grave consequences.
Biomedical datasets can be uncertain and incomplete, with missing, noisy, dirty, and unwanted data. What’s more, some problems in the medical domain can be too complex for a fully automated approach. So, while Machine Learning is often focused on taking humans out of the equation, (driverless cars anyone?), Machine Learning in the healthcare field tends to take a more interactive approach, with domain experts being able to interact with the data, thus working in “partnership” with ML. Two methodologies that work together to support this approach are human-computer interaction (HCI) and knowledge discovery/data mining (KDD). This is known as the HCI-KDD approach.
Interactive Machine Learning or ‘IML’ is defined as the process of building machine learning models iteratively through end-user input. More specifically, it’s defined as “algorithms that can interact with agents and can optimize their learning behavior through these interactions, where the agents can also be human.”*
Interactive Machine Learning makes use of the following styles of learning algorithms
-Reinforcement Learning
-Active Learning
-Online Algorithms**
Indeed, having a “human-in-the-loop” can allow complex problems to be solved more efficiently in many areas of healthcare:
-Health Informatics is one area where the human-in-the-loop is often required. Examples are genome annotation, image analysis, knowledge-base population, and protein structure.
Most of us in the healthcare IT field are familiar with the concept of Artificial Intelligence, or AI. AI attempts to harness the ability of computers and machines learn and perform tasks in order to automate and increase the efficiency of business processes. AI can be grouped into three separate areas of study: deep learning, machine intelligence, and machine learning. Here I will focus on the last of these, machine learning.
Machine Learning automates the analysis of data and inputs to develop computing models needed to operate an automated computer, software program or process, thus eliminating the need to do this manually, saving time, energy, and resources. Sounds great, right? It’s easy to embrace this technology as the way of the future, a means of achieving more efficient operations. However, it’s important to understand the limitations and risks associated with this technology.
-Machine Learning is susceptible to errors and when errors do occur, it can be more difficult to diagnose and correct them.
-Time and accurate data sets are needed to make predictions using Machine Learning, as “learning” is achieved through historical data.
-For scenarios that are not included in the historical data, it is difficult to prove beyond doubt that the predictions made by the Machine Learning system are applicable.
-Machine Learning cannot always provide rational reasons for a particular prediction or decision, as an incomplete or poorly written algorithms do not understand context and are prone to hidden and unintentional biases, depending on the data they have to work with. Furthermore, since Machine Learning is intended to answer questions and not ask them, human collaboration is important in order to better evaluate the output.
This brings us to the question of how Machine Learning can best be applied to Healthcare, a field where mistakes, poor judgement, and inefficient use of time can have grave consequences.
Biomedical datasets can be uncertain and incomplete, with missing, noisy, dirty, and unwanted data. What’s more, some problems in the medical domain can be too complex for a fully automated approach. So, while Machine Learning is often focused on taking humans out of the equation, (driverless cars anyone?), Machine Learning in the healthcare field tends to take a more interactive approach, with domain experts being able to interact with the data, thus working in “partnership” with ML. Two methodologies that work together to support this approach are human-computer interaction (HCI) and knowledge discovery/data mining (KDD). This is known as the HCI-KDD approach.
Interactive Machine Learning or ‘IML’ is defined as the process of building machine learning models iteratively through end-user input. More specifically, it’s defined as “algorithms that can interact with agents and can optimize their learning behavior through these interactions, where the agents can also be human.”*
Interactive Machine Learning makes use of the following styles of learning algorithms
-Reinforcement Learning
-Active Learning
-Online Algorithms**
Indeed, having a “human-in-the-loop” can allow complex problems to be solved more efficiently in many areas of healthcare:
-Health Informatics is one area where the human-in-the-loop is often required. Examples are genome annotation, image analysis, knowledge-base population, and protein structure.
-In clinical medicine, large sets of training data are often not available, as with rare diseases or with malfunctions of humans or machines. What’s more, in the case of emergency medicine or intensive care, time is a major factor, as a healthcare provider may need results in real or a very short time. In such situations, current machine learning methodologies are often insufficient.
-On the flip side, in domains with with increasingly large and complex data sets, the challenge of trying to keep up, or manage “Big Data” makes the creation of integrative and interactive machine learning solutions a must.
-Robotic Surgery is a common example of machine learning with a human-in-the-loop. Complex robotics in tandem with machine learning algorithms allow surgeons to manipulate surgical tools precisely, in order to perform surgery with fine detail, in tight spaces, and with fewer tremors than would be possible for the human hand working alone.
-Medical diagnostics and clinical decision making are complicated processes that involve a multitude of factors. Machine learning algorithms are increasingly being utilized to improve accuracy and efficiency in these functions. A well created set of diagnostic and decision support systems can serve as an extension of scientific knowledge, and assist physicians in making iteratively more precise decisions regarding diagnosis and treatment of disease.***
-Particularly with regard to patient admission, humans have the advantage of being able to see the whole picture at a glance. Our ability to apply creativity, deductive thinking and to apply knowledge from seemingly unrelated situations to another gives us an edge over machines, which cannot be underestimated.
While it’s difficult to negate the benefits of Interactive Machine Learning, thoughtful and evidence-based progress in this field will be essential in improving the process. Especially for those of us in Healthcare IT, it will be exciting to see the new applications of this technology in the field, and to see where improvements in Machine Learning will take healthcare in the decades to come.
--
Kelly Buckman is a healthcare IT expert and field expert blogger for Barracuda Consulting.
Kelly has almost a decade of experience as a Technical Support Engineer/ Analyst in the field of Healthcare IT, over 20 years in IT Support, and several years of experience in Project Management. She has a B.A. from Mount Holyoke, Masters degree from UMass Amherst, and lists her skills as the ability to analyze and resolve various types of application, server and network issues, and to communicate complex ideas effectively.
She is also the mother of 3 sons, ages 19, 17, and 11, lives in western Massachusetts, and enjoys solving puzzles, reading, and travelling.
Please leave your comments below. If you would like to subscribe to our newsletter, click here: https://tinyletter.com/barracuda-consulting. To purchase a full report on this subject, or to access our complete suite of healthcare, and IT advisory services please contact us: https://www.barracuda-consulting.net/contact.
-On the flip side, in domains with with increasingly large and complex data sets, the challenge of trying to keep up, or manage “Big Data” makes the creation of integrative and interactive machine learning solutions a must.
-Robotic Surgery is a common example of machine learning with a human-in-the-loop. Complex robotics in tandem with machine learning algorithms allow surgeons to manipulate surgical tools precisely, in order to perform surgery with fine detail, in tight spaces, and with fewer tremors than would be possible for the human hand working alone.
-Medical diagnostics and clinical decision making are complicated processes that involve a multitude of factors. Machine learning algorithms are increasingly being utilized to improve accuracy and efficiency in these functions. A well created set of diagnostic and decision support systems can serve as an extension of scientific knowledge, and assist physicians in making iteratively more precise decisions regarding diagnosis and treatment of disease.***
-Particularly with regard to patient admission, humans have the advantage of being able to see the whole picture at a glance. Our ability to apply creativity, deductive thinking and to apply knowledge from seemingly unrelated situations to another gives us an edge over machines, which cannot be underestimated.
While it’s difficult to negate the benefits of Interactive Machine Learning, thoughtful and evidence-based progress in this field will be essential in improving the process. Especially for those of us in Healthcare IT, it will be exciting to see the new applications of this technology in the field, and to see where improvements in Machine Learning will take healthcare in the decades to come.
--
Kelly Buckman is a healthcare IT expert and field expert blogger for Barracuda Consulting.
Kelly has almost a decade of experience as a Technical Support Engineer/ Analyst in the field of Healthcare IT, over 20 years in IT Support, and several years of experience in Project Management. She has a B.A. from Mount Holyoke, Masters degree from UMass Amherst, and lists her skills as the ability to analyze and resolve various types of application, server and network issues, and to communicate complex ideas effectively.
She is also the mother of 3 sons, ages 19, 17, and 11, lives in western Massachusetts, and enjoys solving puzzles, reading, and travelling.
Please leave your comments below. If you would like to subscribe to our newsletter, click here: https://tinyletter.com/barracuda-consulting. To purchase a full report on this subject, or to access our complete suite of healthcare, and IT advisory services please contact us: https://www.barracuda-consulting.net/contact.
Comments
Post a Comment