Construction of an AI-assisted teaching platform with multimodal data for tinnitus differential diagnosis

Objective To construct an artificial intelligence-assisted diagnosis teaching platform integrating multimodal data with tinnitus differential diagnosis as the application scenario, for providing the digital tools for clinical teaching in otorhinolaryngology head and neck surgery.

Methods The AI-assisted diagnosis teaching platform was constructed by the differential diagnosis of tinnitus for core and integrating multimodal clinical data such as pure tone audiometry, tinnitus matching, imaging and medical history text. Sixty medical students majoring in clinical medicine of the 2021 grade in Anhui Medical University were selected, and grouped by random number method. The baseline characteristics of two groups were balanced. The experimental group conducted teaching using this platform, while the control group adopted the traditional "theoretical lecture + case analysis" mode. The application value of platform was verified through platform performance tests, teaching effect evaluations and questionnaire surveys.

Results The platform had a fast overall response speed and good stability. The overall diagnostic accuracy rate of the AI diagnostic model for the causes of tinnitus was 86.22%, the sensitivity was 85.41%, and the specificity was 94.56%. The practical assessment score of tinnitus differential diagnosis in the experimental group (87.96 ± 3.18 points) was significantly higher than that in the control group (71.07 ± 4.53 points) (t = 16.80, P ＜ 0.01), while there was no statistically significant difference in the theoretical assessment scores between two groups (P ＞ 0.05). The overall satisfaction rate of the experimental group with the platform teaching reached 93.3%.

Conclusions The platform can effectively enhance the trainees' ability to interpret multimodal data of tinnitus and their level of differential diagnosis, provide a new path for digital teaching in otorhinolaryngology head and neck surgery, and also offer a reference for the construction of AI-assisted diagnosis teaching platforms in other specialties.