Speech Recognition Brings Breast Imaging Report Errors

Speech recognition software produces eight times as many errors as conventional dictation transcription in breast imaging reports, according to new research published in the October American Journal of Roentgenology.

Lead author Sarah Basma of Women’s College Hospital in Toronto, Canada, and colleagues considered 615 breast imaging reports from January 2009 to April 2011. The reports, from two hospitals, were evenly split between those created through automated speech recognition and conventional dictation transcription. They found at least one major error in 23 percent of reports done via speech recognition. With dictation transcription, the rate was 4 percent.

Major errors included word omission, word substitution, nonsense phrases, and punctuation errors, among others.

Errors varied by report type. Breast MRI reports were most prone to them, with 35 percent of speech recognition versions having a major error, 13 percent of interventional procedures, and 15 percent of mammography reports (the dictation equivalents had error rates of 7 percent, 4 percent and 0 percent, respectively).

Seniority and native language had little bearing on error rates, the researchers found.

“We thought that there may be a higher error rate for non-native English speakers because the software works with voice recognition, but that didn’t happen,” said co-author Anabel Scaranelo, MD, of the University Health Network in Toronto.

After adjustment for academic rank, native language, and imaging modality, reports generated with speech recognition were eight times as likely as conventional dictation transcription reports to contain major errors.