AI Prognostic Systems, Gender Differences in Radiology, and More News

First AI prognostic system can predict mortality from chest x-rays

A new study published in JAMA Network Open on July 19, 2019, demonstrated that it may be possible to extract long-term mortality information from chest x-rays using an AI algorithm. Previous studies have shown that AI has great potential for diagnostics, but prognostics remain a challenge.

Given that chest radiographs are the most common diagnostic imaging test given to patients, researchers from Massachusetts General Hospital, Harvard Medical, and Germany’s Stralsund University of Applied Sciences developed a convolutional neural network called CXR-Risk to see whether the tool could offer long-term prognostic information about mortality.

The researchers trained CXR-Risk on more than 85,000 chest x-rays from over 40,000 patients in order to best learn the features on a chest radiograph image that might predict overall health and mortality. When the researchers tested CXR-Risk on 16,000 from previous clinical trials, they discovered that 53% of the patients that the AI designated as “very high risk” died within a 12-year-period and more than 95% of those labeled as “very low risk” survived beyond that time period.

The authors conclude that AI tools can help provide doctors with some prognostic information from a single chest x-ray so they can better identify patients who will most benefit from different lifestyle interventions. In addition, with the addition of other risk factor information to the model, the authors believe they can increase its accuracy to help inform clinical decision making in the future.

Gender differences in interventional radiology

As medicine seeks to remedy its diversity issues, researchers from NYU Langone Medical Center, Albany Medical College, Weill Cornell Medical College, and the Emory University School of Medicine investigated how gender may influence practice patterns in interventional radiologists. Their findings were published in the American Journal of Roentgenology in July 2019.

Interventional radiology, historically, is a field dominated by men. In order to see how women were faring in this discipline, the researchers examined billed clinical work effort of interventional radiologists in a database of Medicare claims data from 2016, taking a close look at all Medicare Part B line item claims for those clinicians. They discovered that only just over 8% of interventional radiologists were women, with the highest percentages of female interventional radiologists located in the western and northeastern United States. In addition, most of the female interventional radiologists were within their first decade of practice.

When the researchers looked at practice patterns in those claims, including service categories, procedural complexity, and patient complexity, they found that women and men displayed similar levels of ability and experience. That said, female interventional radiologists were slightly less likely to make claims for invasive services and non-invasive diagnostic imaging-yet they spent more time than their male counterparts on the evaluation and management for clinical visits.

The authors conclude that women remain underrepresented in the field of interventional radiology.  Yet, because their skills and abilities rival those of their male colleagues, providing key contributions to the field, the researchers hope that this specialty will work harder to address potential biases against female practitioners in the future.

Using MRI features to better clear margins before breast cancer surgery

During breast conserving surgery, surgeons work hard to “clear” the margins, or make sure that no cancer cells remain along the edge of the removed tumor. In doing so, they reduce the risk of the cancer spreading, and any future surgeries. Unfortunately, because cancer cells are very difficult to detect, surgeons may end up leaving positive or close margins in patients. Now, new research published in the European Journal of Radiology suggests there are specific features that can be seen on pre-operative MRI scans that may help reduce repeat surgical procedures.

Researchers from Memorial Sloan Kettering Cancer Center reviewed the pre-operative MR scans of 249 patients who had been diagnosed with invasive ductal carcinoma who later had margins defined as positive, close, or negative. Using multivariate logistic regression analysis, they hoped to uncover imaging and clinical factors that might better predict which cases would end up with positive or close margins.

The study authors found that several imaging features which helped to better predict cases that would have positive or close margins post-surgery. These included multifocal disease, greater background parenchymal enhancement, larger lesion size, and the presence of ductal carcinoma in situ from a needle biopsy.

The researchers concluded that radiologists could help identify these features prior to surgery, helping surgeons to be more aggressive with tumor removal during breast-conserving surgical procedures, and reducing the need for future surgeries.

Novel PET agent can better differentiate between tumors and inflammation

PET imaging can be a useful diagnostic tool for identifying lung cancer, however, high levels of inflammation in the organ can often get in the way of a good read. Since inflammation is a common symptom involved with a variety of medical conditions, finding new ways to differentiate between cancerous lesions and inflammation could help radiologists avoid both false-positives and false-negative results during image evaluation. Researchers from China’s Zhongshan Hospital have now tested a new site-specific imaging agent, ¹⁸F-labeled tracer ¹⁸F-PTTP (5-{[2-Chloro-3-(trifluoromethyl)phenyl]carbonyl}-1-pyrimidin-2-yl-4,5,6,7-tetrahydro-1H-[1,2,3]triazolo[4,5-c]pyridin) for targeting P2X7Rs to see whether it can do the trick. The results were published in the July 2019 issue of the Journal of Nuclear Medicine.

After evaluating the stability and safety of the tracer both in vitro and in vivo, the researchers tested the ¹⁸F-PTTP on both A549 tumor-bearing mice and mice that simply had inflammation. They found the radioligand:

• Showed excellent stability in 0.9% NaCl and 0.1% bovine serum albumin, an affinity to RAW264.7 cells, and rapid blood clearance in mice;

• Demonstrated a peak of uptake in the inflammation mouse model approximately 5 minutes after injection and stayed at an imageable level for 30 minutes;

• And did not show any significant radioactivity uptake in tumor grafts until 60 minutes after injection.

With such differences in uptake, the researchers concluded that ¹⁸F-PTTP has great potential to help radiologists differentiate between solid tumors and inflammation in the lung, as well as aid in the screening of new drugs and the quantification of peripheral inflammation in basic science and clinical studies in the future.

Mini-MRI may help better diagnose knee ailments

To date, it is difficult for clinicians to adequately image the ligaments and tendons in the knee using MRI scans. If clinicians could better image that connective tissue, they could better identify knee injuries and help millions of people who experience knee problems.  Researchers from Imperial College London have developed a “mini-MRI” scanner using the magic angle effect to better diagnose knee injuries in the future. The scientists published a proof-of-concept study in the September 2019 issue of Magnetic Resonance in Medicine.

The collagen fibers that make up the connective tissue in the knee are difficult to see in MRI images because they hold water molecules in such a tight configuration. But researchers from Imperial College London believed if they could develop a scanner that could image the knee at the right, or “magic” angle of 55 degrees, the resulting image would be much brighter and easier to read. The team developed a new prototype scanner that can be pulled over the knee like a metal bracelet, allowing them to easily change the orientation of the magnetic field in order to achieve the angle required to better visualize the knee’s ligaments and tendons.

The researchers tested the prototype on 6 caprine and 10 canine knees. They were scanned by a traditional MRI scanner as well as assessed by a specialized orthopedic veterinarian who offered a diagnosis after dissecting and photographing the joints. The researchers discovered that the method could be used to see partial ligament tears, as well as to visualize age-related changes in collagen structure, that were missed by a normal hospital-grade scanner.

The study authors conclude that while this is an early stage proof-of-concept study, this kind of technology has great potential to accurately detect hard-to-visualize knee injuries in the future.