A study published on June 22, 2026, in the Journal of Clinical Medicine compared the effectiveness of a large language model and a neural network in classifying periodontitis. The retrospective research utilized clinical data from 110 patients, including variables such as age, smoking status, and pocket depth. Experts established the reference diagnoses for evaluation.

The neural network outperformed the LLM, achieving 85% accuracy for stage and 79% for grade classification, with Cohen’s kappa coefficients of 0.79 and 0.67. In contrast, the LLM reached 62% accuracy for stage and 63% for grade, with a kappa of 0.48. Error patterns differed significantly; the LLM tended to underestimate disease severity, while the neural network showed a bias toward overestimating progression. The researchers reported that these performance variations were statistically significant with a p-value below 0.0001.

A study published on June 22, 2026, in the Journal of Clinical Medicine compared the effectiveness of a large language model and a neural network in classifying periodontitis. The retrospective research utilized clinical data from 110 patients, including variables such as age, smoking status, and pocket depth. Experts established the reference diagnoses for evaluation.

The neural network outperformed the LLM, achieving 85% accuracy for stage and 79% for grade classification, with Cohen’s kappa coefficients of 0.79 and 0.67. In contrast, the LLM reached 62% accuracy for stage and 63% for grade, with a kappa of 0.48. Error patterns differed significantly; the LLM tended to underestimate disease severity, while the neural network showed a bias toward overestimating progression. The researchers reported that these performance variations were statistically significant with a p-value below 0.0001.