How AI is helping to unlock the full potential of man and machine in radiology
By Sham Sokka
Today, the quantity, diversity, and complexity of imaging use and the associated data is staggering. Specifically, problem-solving and learning applications driven by artificial intelligence (AI) are already making their mark in patient care. Clinicians in many practice areas are starting to see its benefits—ranging from improved workflow to enhanced clinical decision support—aimed at improving the quality of care.
In radiology, AI is beginning to show its worth as a way to support radiologists and help them be more productive, efficient, and focused on their patients. A tremendous amount of attention is being paid to the clinical uses of AI, but it is still in its infancy for radiology operations and services, where it promises to have just as great an impact.
While the application of AI is evolving, it is capable of providing both clinicians and healthcare technology management (HTM) professionals with the predictive informatics and intelligence that will redefine and enhance how these professionals interpret and share data.
AI Starts to Soar
AI-driven solutions can enable radiology departments to become stronger and more productive than ever before, with more visibility into their operational issues, from equipment maintenance to scheduling. It all starts with data. Data provides HTM professionals with more than just a gut feeling.
Data delivers actual information and insights into what is happening—and through AI and predictive analytics, offers further foresight to better respond to what is likely to happen.
Still, these capabilities do not stand to replace the responsibilities of clinical or technology professionals. What they do, however, is provide these individuals with greater information so that they may work more efficiently, thus improving operations and enhancing the patient experience.
There is that subconscious “gut feeling” technology managers have from years of experience, combined with integration of many disparate factors into a complex, pattern matching. In many ways, the more sophisticated forms of AI attempt to mimic this “gut feeling.”
The Benefits of AI
AI can provide clinicians with the means to do even more extraordinary things. In operations, deep learning technologies can make significant improvements in everything from scheduling optimization to foreknowledge of equipment downtime.
For example, about 10%-12% of all radiology appointments are no-shows or last-minute cancellations. This creates big problems for daily schedules and contributes to lost revenue. Today, AI-based technology can predict whether a particular patient is going to show up for his or her appointment with approximately 90% accuracy.
With this information in-hand, departments can call the anticipated no-show patients to provide reminders or double-book that time slot; thus, AI enables scheduling to run more smoothly and help ensure revenue capture.
Furthermore, from an equipment maintenance standpoint, AI can provide tremendous value. By collecting data from all imaging machines, HTM professionals can better predict if a system is going to have unplanned downtime or may otherwise need support. For example, technology managers may find that every three months a machine consistently displays the same error and slows employee workflow, although a regular software patch update can prevent it.
Plus, having the right data can enable teams to predict such an occurrence and issue patches proactively in advance, to avoid downtime, maintain workflow, and provide a better experience for the staff and their patients. In other words, AI makes zero-unplanned-downtime a possibility.
However, while there is a lot of excitement around data insights and predictive analytics, it is the practical application that will determine overall capabilities and benefits. For hospitals and HTM professionals looking to integrate AI and machine learning technologies into their current radiology operations, here are three key considerations:
1. Accurate Data Collection: The first step to using AI technology is collecting data and making sure that it is complete, clean, and organized with a consistent nomenclature. This gives healthcare technology managers a comprehensive picture of radiology operations. For instance, one must monitor the temperature centers around all of the imaging equipment to ensure that the devices are functioning properly.
Since heat is a good indicator of a possible system failure, AI-driven data insights can help radiology departments or equipment providers predict if a system may experience problems. That way, they can take proper action before issues arise.
2. Standardization: We are all familiar with the expression: “Bad data input equals bad data output.” Artificial intelligence, like human intelligence, operates best with high-quality information. When making decisions that may affect how a healthcare organization operates and cares for patients, it is important to ensure that decision-making is based on sound interpretation of the best possible data.
Standardization is a good strategy to ensure accuracy, consistency, and totality of data. And employing consistent processes to obtain and label data allows for precision.
Further, accuracy occurs through multiple iterations or detect-correct cycles, eliminating untrustworthy data. What’s key is the completeness of data, although most AI systems can handle some sporadic missing information. However, standards in labeling data are necessary to enable AI techniques, as most algorithms draw correlations and inferential logic to associate different variables.
While we recognize that in the real world no data is perfectly accurate, consistent or complete, it’s promising to note that AI methods can also tolerate a level of imperfection that is realistic. In fact, one can even employ these methods to detect imperfections in the data themselves and to suggest strategies to improve data acquisition. After all, once data is standardized, AI can pay handsome dividends.
3. Change Management: Implementing any kind of change—particularly new technology—into a busy clinical workflow can be a challenge. That’s why it’s critical to implement workforce training to help hospital staffers understand how to use and interpret data in new ways. Beyond using the data and outputs to make a significant impact on day-to-day operations, it is important to explain to staff why the changes are being made and how they will help to evolve operations.
Radiology professionals, in particular, must have a clear idea about how the data might be translated into practice—what it could look like and how it will impact the way they work. Helping the staff to understand the technology, data, and goals will encourage them to embrace AI-driven solutions and know how they will benefit the entire practice and its patients in the long run.
Still, AI is not meant to replace personnel—but to help them perform higher-level tasks that demand human ingenuity, creativity, and compassion.
Making Sense of AI
To sum it up, by implementing AI technologies with these three focuses, HTM professionals are now positioned with a 360° view of their department and all the technology and equipment they support. Such transparency then equips them with the necessary information to predict operational issues before they even arise. New AI and deep learning innovations are also enabling them to be more proactive and keep systems running optimally.
Even so, we are just beginning to scratch the surface with AI in healthcare operations. What’s happening today will transform healthcare tomorrow—and it all starts with data and people. Through a combination of strong leadership and staff willingness, healthcare technology managers will have the opportunity to do more extraordinary things for their radiology departments.
Sham Sokka is general manager of radiology solutions at Philips. For more information, contact 24×7 Magazine chief editor Keri Forsythe-Stephens at [email protected].
AI is probably over optimistically named. We might also get Artificial Mediocrity if not Artificial Stupidity. (With apologies to those who have read these words before.) It is also hardly new, with the forerunner “Expert Systems” having been much discussed in the 1970’s and 80’s, and then slipping down into the mire of the over hyped.