Exploration Of Ai Large Models In Orthodontic Treatment Planning

To compare the consistency of the orthodontic treatment plans designed by the AI large models and the orthodontists. A case of an 18-year-old patient with skeletal Class III malocclusion (ANB angle -3.2 degrees) and anterior crossbite was selected. The patient’s upper and lower midlines are both shifted 2mm to the right, with the crown of the upper right first premolar rotated 60 degrees mesially, and severe crowding in the upper left arch. The patient's dental alignment, facial profile, and radiographic findings were input into DeepSeek-R1 and ChatGPT 4.0 to generate orthodontic camouflage treatment plans. Simultaneously, ten orthodontists were asked to formulate orthodontic camouflage treatment plans based on the patient's examination data. The consistency of the orthodontic treatment plans generated by the two AI models and those proposed by the ten orthodontists was evaluated. Among the ten orthodontists, two proposed an extraction treatment plan. Both DeepSeek-R1 and ChatGPT 4.0 also selected extraction plans, with a concordance rate of 20%. The concordance rate between DeepSeek-R1 and the orthodontists regarding the number of extracted teeth was 10%, while ChatGPT 4.0 had a concordance rate of 0%. Neither AI model demonstrated concordance with the orthodontists regarding the specific extraction sites, with a concordance rate of 0%. There are significant differences between AI large models and orthodontists in orthodontic treatment planning, with additional discrepancies observed between the DeepSeek-R1 and ChatGPT 4.0 models themselves.