April 10, 2025 | Researchers worldwide are looking to revamp traditional approaches to medical imaging and develop alternative solutions that could improve visualization, speed, and accuracy in diagnostic settings.
At the University of Arkansas, for example, one group is developing an AI tool designed to help radiologists and doctors analyze chest X-rays and computed tomography (CT) scans, with an emphasis on understanding how the AI assistant arrives at its conclusions.
“One of the biggest challenges in AI for healthcare is the lack of explainability, which makes it difficult for doctors to trust automated decisions,” said Ngan Le, assistant professor of computer science and computer engineering and director of the Artificial Intelligence and Computer Vision Lab at the University of Arkansas.
Their new development “bridges this gap by providing interpretable decision-making, allowing radiologists and physicians to understand and validate AI-driven insights,” she told Diagnostics World.
Meanwhile, technology pursued by University of Houston researchers can separate X-ray photons and help identify specific materials or contrast agents within the body. According to the group, their approach may help improve cancer detection, among other possible benefits.
In another line of work, a group at Kindai University in Japan is building a system to monitor patient respiration and breath-holding during diagnostic imaging procedures.
As he envisions the future potential, Hiroyuki Kosaka told Diagnostics World that one notable upside is the ability to perform safe and accurate exams for patients, such as infants or older adults, who may not be able to follow breath-holding instructions.
We took a closer look at these new developments, their potential, and some of the challenges that will need to be addressed before these approaches can be used in real-world settings.
AI Tool Designed to Read X-Rays Like Radiologists
In a paper published in the journal Artificial Intelligence in Medicine (DOI: 10.1016/j.artmed.2024.103054), the University of Arkansas’s Le and co-authors describe their tool, named ItpCtrl-AI (interpretable and controllable artificial intelligence), which they refer to as a novel framework “that mirrors the decision-making process of the radiologist.”
The researchers studied where radiologists looked and how long they focused on certain areas while examining chest X-rays. The heatmap formed from that eye-gaze dataset helped train the computer on where to search for concerning abnormalities and learn which image areas tend to need less time and attention.
“Our approach stands out for its ability to generate anatomical attention heatmaps and predict abnormal findings in chest X-ray images,” they wrote. “By leveraging anatomical prompts, our model offers a unique layer of control and flexibility, allowing users to specify the type of heatmap and diagnosis they seek.”
To the best of the group’s knowledge, their method is “the first in the medical domain to learn from radiologist-based anatomical gaze attention while offering controllability.” Upon acceptance, they indicated plans to release the trained models, source code, and annotated dataset.
When asked what sets this apart from other medical imaging solutions, Le emphasized the ability to understand how the technology arrives at its conclusions and highlighted the importance of that ability.
“Unlike traditional black-box AI solutions, our approach prioritizes interpretable decision-making, ensuring transparency and explainability in medical imaging analysis,” she said. “This allows healthcare professionals to understand and trust the AI’s outputs, leading to more reliable and accountable clinical decisions.”
Le’s group is now working to refine ItpCtrl-AI so that it can read more complex 3D scans. CT scans provide far richer information than 2D chest X-rays, she explained, but processing them is also more complicated. Beyond that, there is no existing gaze data specifically for CT scans, so that presents another significant hurdle.
To address these challenges, she and colleagues have gathered a dataset at the MD Anderson Cancer Center that includes gaze data, CT scans, medical reports, and findings. They have been working on preprocessing the data and plan to expand their dataset moving forward.
The possibility of scaling this AI tool beyond CT scans excites Le as she considers the future potential of ItpCtrl-AI and its possible real-world applications. She also noted hopes of integrating ItpCtrl-AI into real-world clinical workflows to improve workflow efficiencies and diagnostic accuracy.
“Our short-term goal is to develop an interpretable AI-driven decision-making system that can assist radiologists and doctors in reading and analyzing CT scans effectively,” she added. “Looking ahead, our long-term vision is to generalize this approach to support medical professionals in interpreting a broad range of medical imaging data across various modalities.”
Photon Counting Detectors Could Improve Visualization
Medical professionals have traditionally relied on 2D X-rays to diagnose issues like bone breaks, but sometimes other issues like soft tissue damage can be overlooked along the way. And alternatives like MRI scans, which take longer and carry a higher price tag, aren’t always an option. University of Houston researchers are pursuing a new 3D approach that may help address this dilemma.
“Right now, X-rays used in medical clinics and other industries collect incoming photons as a whole, similar to how white light contains all the colors, but they aren’t separated,” said Mini Das, a professor at the University of Houston’s College of Natural Sciences and Mathematics and Cullen College of Engineering. “So, while they can show differences in density – like distinguishing between bone and soft tissue – they can’t tell us exactly what materials are present.”
Photon counting detectors developed by her research group can separate X-ray photons by energy level and help identify certain materials or contrast agents used in medical imaging. According to Das, this could potentially improve the detection of cancer.
“If you inject two different contrast agents – one targeting a tumor and another targeting inflammation – you could see where each one accumulates,” she explained. “Right now, we can see bright areas in an image, but we can’t always tell what they are. This technology would give us a much clearer, quantitative analysis. It would allow us to determine not just what’s inside an object, but what different materials are present and in what quantities.”
The researchers envision a wide array of possible applications for their technology, which extends beyond medical imaging into areas such as materials imaging, baggage scanning for security purposes, and micro-electronics and nano-electronics imaging.
Das and colleagues are in the research and development phase at this point and are currently working with industry collaborators in Europe to develop larger versions of the new detectors and improve their measurement accuracy and performance. After those challenges are addressed, testing could begin in real-world settings.
This line of research fits within a bigger picture of research taking place in this space, and Das recently co-authored an editorial paper that introduces a special issue devoted entirely to photon counting detectors in the Journal of Medical Imaging (DOI: 10.1117/1.JMI.11.S1.S12801).
She and the University of Chicago’s Patrick J. La Riviere detailed the advantages photon counting detectors offer over traditional energy-integrating detectors, acknowledged the variety of challenges faced by photon counting detectors despite their promise, and highlighted an array of research papers ranging from fundamental questions about photon counting detector operations to improved optimization and quantification, among other topics.
In recent years, photon counting detectors have transitioned from research labs to clinical prototypes, they wrote, and are now “on the verge of FDA clearance” for CT scanners. “It seems inevitable,” they concluded, that photon counting detectors “will soon find widespread use in clinical and industrial imaging and these papers provide a timely snapshot of this exciting transition.”
Respiratory Motion Monitoring System Aims to Enhance Image Quality
Researchers in Japan have developed a respiratory motion monitoring system for X-ray imaging and CT scans. Their millimeter-wave sensor (MWS) technology uses electromagnetic radiation to detect motion during diagnostic imaging procedures without any physical contact, which they say helps with patient comfort and privacy.
Among other advantages of this new system, the researchers highlighted its ability to maintain accuracy through clothing and provide stable measurements in both standing and supine positions. In addition, they say their solution is cost-effective compared to existing technologies and can be easily integrated with current X-ray and CT equipment.
“This technology has the potential to standardize respiratory monitoring across diagnostic imaging,” said Kindai University’s Hajime Monzen. “By providing objective, real-time feedback, we can significantly reduce the need for repeat imaging and improve diagnostic accuracy.”
Respiratory monitoring is needed for the confirmation of breath-hold and for four-dimensional CT reconstruction, according to Monzen and colleagues. Whereas respiratory monitoring systems are commonly used in radiation therapy, they are not frequently used for diagnostic imaging purposes, where they could improve image quality.
With this in mind, the researchers pursued a study that used their sensor to “non-invasively visualize respiratory motion, confirm breath-holding, and explore the potential for clinical implementation of an MWS in diagnostic x-ray imaging, CT, and radiation therapy.” Their findings appeared in Medical Physics (DOI: 10.1002/mp.17616). The MWS technology “successfully monitored respiratory motion and breath-holding during radiographic and CT imaging,” they concluded.
The novelty of this study is “the successful visualization of the patient’s respiratory status and breath-holding at the same distance as in clinical practice . . . using 24 GHz microwaves,” Kindai University’s Hiroyuki Kosaka told Diagnostics World.
He and colleagues used a radio-wave dark-box system to test the MWS and its ability to detect movement from different angles. They also used a technique called the fast Fourier transform to optimize the sensor and learn more about its capacity to identify and separate relevant breathing signals. Comparing breathing patterns with those from a respiratory motion phantom called QUASAR allowed them to simulate breathing with differing levels of motion and confirm that their technology can, indeed, reliably track respiratory motion.
There are limitations to consider, however. For example, “the sensor detects motion at a limited number of points on the target surface rather than mapping motion in three dimensions, like a camera,” the researchers acknowledged. Another limitation is that it cannot quantify a patient’s motion in certain directions because “the MWS waveform can undergo a velocity-dependent frequency shift,” they noted. If these issues can be addressed, they foresee the ability to sort CT projections and correct imaging artifacts, among other clinical applications.
Kosaka indicated that his group is now working with equipment manufacturers to implement this new technology in medical scenarios. Looking ahead, he explained that one of the biggest challenges will be finding the optimal location for the sensor, which involves “taking into account the effective range of the microwaves and the position” to avoid interfering with X-ray exams.
Paul Nicolaus is a freelance writer specializing in science, nature, and health. Learn more at www.nicolauswriting.com.