Radiology in public focus

Press releases were sent to the medical news media for the following articles appearing in recent issues of RSNA Journals.

Fig 4 Erickson

Example of how improper feature removal from imaging data may lead to bias. (A) Chest radiograph in a male patient with pneumonia. (B) Segmentation mask for the lung, generated using a deep learning model. (C) Chest radiograph is cropped based on the segmentation mask. If the cropped chest radiograph is fed to a subsequent classifier for detecting consolidations, the consolidation that is located behind the heart will be missed (arrow, A). This occurs because primary feature removal using the segmentation model was not valid and unnecessarily removed the portion of the lung located behind the heart.

https://doi.org/10.1148/ryai.210290 ©️RSNA 2022

 

Special Report Lays Out Best Practices for Handling Bias in Radiology AI

With increased use of AI in radiology, it is critical to minimize bias within machine learning (ML) systems before implementing their use in actual clinical scenarios, according to a special report published in Radiology: Artificial Intelligence.

The report is the first in a three-part series developed by Bradley J. Erickson, MD, PhD, professor of radiology and director of the AI Lab at the Mayo Clinic, in Rochester, MN, and colleagues. The team outlined suboptimal practices used in the data handling phase of ML system development and presented strategies to mitigate them.

According to the report, developers must accurately handle data in challenging scenarios. Careful planning should include an in-depth review of clinical and technical literature and collaboration with data science expertise.

To develop a more heterogeneous training dataset, the authors suggest collecting data from multiple, geographically diverse institutions, using data from varied vendors and times of day, or including public datasets.

“Creating a robust machine learning system requires researchers to do detective work and look for ways in which the data may be fooling you,” Dr. Erickson said. “Before you put data into the training module, you must analyze it to ensure it’s reflective of your target population. AI won’t do it for you.”

The second and third reports in the series focus on biases that occur in the model development and model evaluation and reporting phases.

For More Information

Access the Radiology: Artificial Intelligence study, “Mitigating Bias in Radiology Machine Learning: 1. Data Handling.”

Fig 6 Tse
 

AI-Based System Shows Promise in Tuberculosis Detection

An AI system detects tuberculosis (TB) in chest X-rays at a level comparable to radiologists, according to a study published in Radiology. Researchers said the system may be able to aid screening in areas with limited radiologist resources.

“We have effective drugs for treating TB, but large-scale screening programs to detect TB are not always feasible in low-income countries due to cost and availability of expert radiologists,” said study co-author Rory Pilgrim, BEng, a product manager at Google Health AI in Mountain View, CA.

Study first author Sahar Kazemzadeh, BS, software engineer at Google Health, and colleagues developed and assessed a deep learning-based AI system that can quickly and automatically evaluate chest X-rays for TB. The researchers developed the system using data from nine countries and tested it on data from five countries, covering multiple high-TB-burden countries, various clinical settings and a wide range of races and ethnicities.

Analysis by 14 international radiologists showed that the system performed well and was comparable to radiologists in determining active TB on chest X-rays.

“What’s especially promising in this study is that we looked at a range of different datasets that reflected the breadth of TB presentation, different equipment and different clinical workflows,” Kazemzadeh said. “We found that this deep-learning system performs really well with all of them with a single operating point.”

For More Information

Access the Radiology study, “Deep Learning Detection of Active Pulmonary Tuberculosis at Chest Radiography Matched the Clinical Performance of Radiologists,” and the related editorial, “Tuberculosis Detection from Chest Radiographs: Stop Training Radiologists Now.”

Fig 5 Baffour

Images in a 71-year-old man with multiple myeloma. Lytic lesions (dashed arrows) within a thoracic vertebral body and the left iliac wing are more conspicuous on the noncontrast-enhanced axial photon-counting detector (PCD) CT image (middle; solid arrows) compared with noncontrast-enhanced axial energy-integrating detector CT image (left). With 0.6-mm Br76 noncontrast-enhanced axial PCD CT reconstruction images (right), more lesions were detected (arrowheads).

https://doi.org/10.1148/radiol.220311 ©RSNA 2022

Novel Photon-Counting CT Improves Myeloma Bone Disease Detection

New CT technology paired with AI-based noise reduction offers superior detection of bone disease associated with multiple myeloma at lower radiation doses than conventional CT, according to a study published in Radiology.

By directly converting individual X-ray photons into an electric signal, photon-counting detector CT can decrease the detector pixel size and improve the image’s spatial resolution. It has demonstrated much better dose efficiency than standard CT, which allows for acquisition of ultra-high-resolution images of large areas of the body.

This potential for improved image quality in whole-body low-dose scans inspired study lead author, Francis Baffour, MD, a diagnostic radiologist at the Mayo Clinic in Rochester, MN, and colleagues to study the technology in people with multiple myeloma, a disease that forms in plasma cells. Bone disease characterized by areas of bone destruction known as lytic lesions is found in approximately 80% of multiple myeloma patients.

The researchers compared photon-counting detector CT with conventional low-dose, whole-body CT in 27 multiple myeloma patients, median age 68 years. The patients underwent whole-body scans with both types of CT and two radiologists compared the images.

They also applied a deep learning AI technique developed at Mayo Clinic’s CT Clinical Innovation Center to reduce the noise in the very sharp photon-counting images.

“We were excited to see that not only were we able to detect these features of multiple myeloma disease activity more clearly on the photon-counting scanner, but with deep learning denoising techniques that allowed us to generate thinner image slices, we were able to detect more lesions than on the standard CT,” Dr. Baffour said.

For More Information

Access the Radiology study, “Photon-counting Detector CT with Deep Learning Noise Reduction to Detect Multiple Myeloma.”

Media Coverage of RSNA

In August, 1,117 RSNA-related news stories were tracked in the media. These stories had over 678 million audience impressions.

Coverage included Yahoo!, Doctor Radio, WGCL-TV (Atlanta), WJZ-TV (Baltimore), U.S. News & World Report, Pittsburgh Post-Gazette, United Press International, MSN.com, Healthline, HealthDay, Medscape, Science Daily, WebMD, Drugs.com, Applied Radiology and Health Imaging News.

 

New Pancreatitis Article on RadiologyInfo.org

Encourage your patients to visit RadiologyInfo.org, the public information website produced by RSNA and the American College of Radiology, for easy-to-read information about the risk factors, diagnostic tools and treatment options for pancreatitis.

RadiologyInfo.org

lung cancer screening compliance feature

November Public Information Outreach Activities Focus on Lung Cancer Awareness

Lung cancer causes more than 350 deaths each day—more than breast, prostate and pancreatic cancers combined, according to Cancer.org.

In recognition of National Lung Cancer Awareness Month in November, RSNA distributed public service announcements to inform patients about the risk factors, available screening methods and treatment options for the disease.