Wearing a mask does cause an increase in report errors with speech-recognition software, but the mistakes are small and unlikely to cause a negative impact.
Wearing a surgical mask while using speech recognition to complete a radiology report does lead to a higher dictation error rate – but, don’t worry. Most mistakes are minor are unlikely to negatively impact care.
Masks have become a necessary and ubiquitous part of healthcare throughout the COVID-19 pandemic, and some in the industry have worried that wearing a mask while using speech-recognition software could open the door for unintended errors.
In a poster presented during the Society for Imaging Informatics in Medicine (SIIM) 2021 Virtual Annual Meeting, Abiola Femi-Abodunde, M.D., a diagnostic radiology resident from the University of North Carolina at Chapel Hill School of Medicine, revealed that masks do cause a 25-percent increase is dictation errors, but they’re small.
“There is a near-significant increase in the rate of dictation errors in unedited radiologist reports created with speech-recognition software, a difference which may be accentuated in some groups of radiologists,” she said. “However, most errors are minor single-word errors and are unlikely to result in a medically relevant misunderstanding.”
For more SIIM 2021 coverage, click here.
For this study, the team created 40 model radiology reports that simulated radiographic/fluoroscopic, ultrasound, CT, and MRI imaging reports that are typically captured in academic medical centers. Six radiologists, who were randomly assigned to masked and unmasked groups, used speech-recognition software to dictate those reports. Their versions were, then, compared to the models, and the team classified any errors by type and severity.
According to their analysis, the overall dictation error rate was 25-percent higher when radiologists wore masks. Specifically, masked providers made 27.1 +/- 2.2 errors per 1,000 words compared with the 21.7 +/- 4.9 errors per 1,000 words from unmasked providers. But, after excluding one trainee with an accented speech pattern, the overall error rate fell to a 19-percent increase while masked – 20.1 +/- 2.2 errors per 1,000 words during masking and 16.9 +/- 1.9 errors per 1,000 words unmasked. Of the unmasked errors, she said, 58 percent were minor, as were 56 percent of those occurring while masked.
CT Perfusion Study Shows Enhanced Detection of Medium Vessel Occlusions with Emerging AI Software
May 21st 2025The Rapid CTP AI software offered 23 percent greater detection of medium vessel occlusions in comparison to the Viz CTP AI software, according to research presented at the European Stroke (Organization) Conference (ESOC).
Can AI Predict Future Lung Cancer Risk from a Single CT Scan?
May 19th 2025In never-smokers, deep learning assessment of single baseline low-dose computed tomography (CT) scans demonstrated a 79 percent AUC for predicting lung cancer up to six years later, according to new research presented today at the American Thoracic Society (ATS) 2025 International Conference.
The Reading Room: Artificial Intelligence: What RSNA 2020 Offered, and What 2021 Could Bring
December 5th 2020Nina Kottler, M.D., chief medical officer of AI at Radiology Partners, discusses, during RSNA 2020, what new developments the annual meeting provided about these technologies, sessions to access, and what to expect in the coming year.
Mammography Study Compares False Positives Between AI and Radiologists in DBT Screening
May 8th 2025For DBT breast cancer screening, 47 percent of radiologist-only flagged false positives involved mass presentations whereas 40 percent of AI-only flagged false positive cases involved benign calcifications, according to research presented at the recent American Roentgen Ray Society (ARRS) conference.