Using artificial intelligence, computers can watch surgeons perform operations and then write post-surgical operative notes that are more accurate than what the doctors themselves would have written, a new study suggests.

Operative notes – reports that document the details of a surgical procedure – are tedious to write and often contain inaccuracies and incomplete information, researchers noted in a report of the study in the Journal of the American College of Surgeons.

Operative reports “not only facilitate communication between healthcare providers, but also serve as the basis for surgical billing and coding, are used for surgical quality benchmarking, enable surgical research efforts, and track compliance with regulatory requirements and evidence-based guidelines," the researchers wrote. "(They) are arguably the single most important document in all of surgery.”

Using AI technology, researchers trained computer-vision systems to detect surgeons’ actions in videos of robotic-assisted operations to remove the prostate.

For each possible step of the operation – for example, lymph node removal, tying off veins, or cutting through the urethra – the researchers pre-wrote descriptive text. As the AI system “watched” the video, it detected the surgeon’s steps and compiled the text into a narrative operative report.

When the researchers tested the system using videos of 158 real-world cases, 53% of reports written by the surgeons contained discrepancies, compared with 29% of AI reports, as determined by an expert team of reviewers.

Significant discrepancies with actions recorded in the videos that could potentially matter to the patient’s subsequent care were found in 27% of surgeons’ reports and in 13% of AI reports.

With further testing, the new technology has the potential “to reduce documentation burden, improve operative report accuracy, promote surgical transparency, and decrease subjectivity in surgical documentation,” the researchers said.

AI BEATS HUMANS AT ANALYZING LONG-TERM HEART MONITOR DATA

AI is better than humans at analyzing long-term heart rhythm monitoring, according to a large study.

The human heart beats up to 120,000 times a day, so analyzing long-term electrocardiograms that may have recorded every heartbeat for days or weeks is a time-consuming process, researchers wrote in Nature Medicine.

Using recordings from 14,606 patients who wore ECG devices for an average of 14 days, the researchers first had the data reviewed by human technicians using standard methods.

They re-analyzed the data using an AI algorithm – “DeepRhythmAI” – developed for the task by Polish company MEDICALgorithmics.

Severe arrhythmias were missed in 4.4% of patients by the human reviewers and in only 0.3% of patients by the AI.

The AI model was able to rule out severe arrhythmia with 99.9% confidence in a 14-day ECG recording, according to the report.

Lack of trained staff to analyze so-called ambulatory ECGs “leads to a huge bottleneck in healthcare worldwide, and at the same time, patients would benefit if we did more and longer ambulatory ECG recordings, not shorter,” study leader Linda Johnson of Lund University in Sweden said in a statement.

“We believe that AI could solve this problem.”