AI Outperforms Dentists in Jaw Scan Analysis, Study Finds
Researchers say the results could reshape diagnostic approaches in dentistry.
Researchers from St. Petersburg and Veliky Novgorod tested artificial intelligence on the analysis of tomographic jaw scans under conditions designed to closely replicate real clinical practice. The results were striking: neural networks proved both faster and more accurate than dentists.
Diagnocat Algorithm Versus Ten Dentists
For the experiment, scientists created three-dimensional jaw models of 100 patients using cone-beam computed tomography. The resulting images were divided into 3,200 anatomical segments. Each segment corresponded either to an individual tooth or to the area where a tooth was missing, allowing researchers to analyze potential pathologies in detail.
The Diagnocat algorithm and ten dentists with an average of about 13 years of professional experience independently analyzed the tomographic data. The results were verified by three independent experts - two dentists and a radiologist with more than 20 years of experience working with dental cone-beam CT.
In the anterior part of the jaw, where the incisors and canines are located, the AI made no diagnostic errors. In the more complex lateral region, accuracy was also higher than that of the dentists. The algorithm made mistakes in 4.86 percent of cases, while doctors made errors in 7.29 percent. The average analysis time for one scan was 4.18 minutes for the AI system, compared with about 25 minutes for specialists. In other words, the algorithm processed tomographic images more than six times faster.
No System Is Perfect
However, the system was not flawless. In about 12 percent of cases, the algorithm failed to detect tooth decay. It also occasionally confused tooth roots with implants and did not always accurately determine the degree of tissue damage.
“Diagnocat’s high sensitivity compared with the results obtained by physicians, along with its high analysis speed, allows the system to be considered an effective tool for supporting clinical decision-making,” said Roman Rozov, professor of prosthetic dentistry at Pavlov First St. Petersburg State Medical University.
Developers emphasize that such systems should not operate autonomously. Instead, the algorithms are viewed as diagnostic support tools, while the final decision and treatment strategy remain the responsibility of the physician.
Earlier reports described how scientists in Volgograd identified a method for early detection of cancer in dentistry.
