AI is pushing the medical imaging industry beyond human limitations

Introduction

According to an academic article published in Nature Partner Journals, “the number of life science papers describing [artificial intelligence (AI)] rose from 596 in 2010 to 12,422 in 2019.” This staggering ~2000 percent increase is primarily a reflection of two exciting developments: one, AI’s quickly growing role in radiology and, two, AI’s capacity to redefine efficiency and accuracy standards in the greater medical imaging industry.

As of September 2020, 74 percent of FDA-approved AI-based medical devices and algorithms were specialized for radiology – and for a good reason. Radiology’s natural abundance of visual data makes the specialty well-suited for early adoption of “smart” technology. AI tools typically require vast volumes of data to run effectively – the more, the better – and while other medical specialties must undertake data collection, aggregation, and organization efforts before exploring AI, radiology benefits from already having it in spades.

This blog explores documented applications of AI in radiology. Additionally, we discuss AI’s role in the future of medical imaging – how the industry will likely evolve alongside technology. Healthcare service providers, particularly those interested in better, faster, more accurate medical imaging, should read on to learn more.

Why Radiology Needs AI

Consider the definition of “radiology”: according to RAD-AID International (a nonprofit by and for radiologists), “a specialty of medicine in which images of the body’s organs are interpreted in order to diagnose disease.”

To those unfamiliar with radiology, a radiologist’s fundamental goal: to efficiently and accurately diagnose disease based on medical images, might seem underwhelming. After all, how hard can it be to “read” a picture? However, this perspective fails to recognize two obstacles standing between radiologists and successful radiology:

  1. Image quality. Disease markers can be tiny, even microscopic. As a result, reading medical images can be like searching for a needle in a haystack. This issue is compounded by the possibility of image quality errors and the fact that organs may vary in size, shape, and even color between individuals. This variance means that radiologists aren’t just looking for a needle in a haystack; they’re looking for a microscopic needle in the world’s most complex haystack – without knowing what the needle looks like.
  2. Workload. The global shortage of healthcare workers exposed by the COVID-19 pandemic will likely increase over time. According to research and consulting firm McKinsey & Co., 2030 will see a shortfall of 9.9 million healthcare professionals, including physicians, nurses, and midwives. The result is more work for fewer workers, and radiology is no exception. One study conducted in a large general hospital in Western Europe found that the average on-call workload for radiologists more than quadrupled between January 2006 and 2020. However, in the same period, the number of radiology staff did not increase.

Symptoms of overworking, including burn-out, eye strain, and stress-related illness, can turn a radiologist’s already difficult task into a near-impossible one. Consequently, radiologists and their sponsor institutions must seek out technology to push radiology beyond the limitations of human capabilities: namely, AI.

Existing AI Applications for Radiology

Already, existing AI applications can not only help aid radiologists by assisting in their work; these tools can also perform tasks beyond the abilities of even the most capable radiologists in the world. 

Harvard Medical School researchers document assistive AI applications in the following areas of radiology:

  • Thoracic imaging. AI can assist in lung cancer screening by detecting pulmonary modules and classifying identified nodules as either benign or malignant. A study published in Clinical Radiology found that partnering radiologists with AI dropped the rate of missed lung cancers on chest X-rays by 60 percent.
  • Abdominal and pelvic imaging. AI can detect and classify liver lesions and other abnormalities.
  • Colonoscopy. AI can detect and classify colonic polyps. 
  • Mammography. AI can detect and classify microcalcifications. This is particularly useful because screening mammography is notoriously tricky; according to an analysis published in Academic Radiology, radiologists miss 24 percent of interval breast cancers on initial screening.
  • Brain imaging. AI can assist in brain cancer screening by detecting abnormal tissue growth and classifying identified growths as benign, malignant, primary, or metastatic.
  • Radiation oncology. AI can automate radiation treatment planning and enhance radiation treatment itself by segmenting tumors for radiation dose optimization. 

Furthermore, expansive AI applications include revolutionary pattern recognition, data analysis, and quantitative imaging. These applications go beyond human capabilities, pushing into uncharted – and promising – territory. 

One recent, exciting example of an expansive AI application in radiology is a fracture-detecting AI developed by the University of Michigan and the Taiwan-based Center for Artificial Intelligence in Medicine. Scientists from these institutions developed an AI model to identify microscopic fractures imperceptible to a radiologist’s eye. Tested on scaphoid breaks (the most common type of wrist fracture), their product was more than 90 percent successful. 

Synaptiq is currently working on a similar project. We are developing an AI model to measure the morphology of arteries based on ultrasound images – an application with huge future potential.

AI and the Future of Medical Imaging

Ongoing testing and research indicate that AI can streamline processes, improve efficiency, and increase accuracy in radiology. Assistive AI applications give radiologists “superpowers” by simplifying their work – automating routine tasks and helping with image screening. Further, expansive AI applications are beginning to challenge long-standing efficiency and accuracy standards in the greater medical imaging industry.

At present, medical imaging necessarily occurs on a 2D level. For example, MRI scans take 2D “slices” of an organ because humans cannot understand and, consequently, cannot make a diagnosis based on 3D representations.  However, AI can understand 3D data. Unlike a human being, an AI model can track every relevant vector in an organ’s living, moving tissue, allowing for a more accurate and holistic diagnosis. 

Ultimately, the future of medical imaging is expansive AI applications that specialize in (1) analyzing data that humans find difficult or impossible to visualize, such as raw MRI data, and (2) triangulating data from multiple sources. Radiologists may have difficulty cross-referencing data generated by the multitude of diagnostic tests available today: blood tests, ECGs, physical examinations, et cetera. By contrast, AI can conduct comprehensive analyses and pattern recognition almost instantaneously, making it a truly revolutionary tool.

About Synaptiq

Founded in 2015, Synaptiq is a data science and AI consultancy with over 50 clients in more than 20 sectors worldwide. We develop and deploy actionable solutions using machine learning, machine vision, natural language processing, and other data-driven techniques. We help clients discover, organize, and leverage the data they have to streamline processes, increase productivity, and drive further innovation.

For more information about Synaptiq, please visit www.synaptiq.ai

Additional Reading