Enhancing AI Diagnosis of Congenital Auricular Deformities With Multicenter Data Sets

OBJECTIVE: The study aimed to address the challenges of early diagnosis of congenital auricular deformities in children by leveraging artificial intelligence (AI), particularly convolutional neural networks (CNNs). Despite AI's potential for improving diagnostic precision, its clinical use has been constrained by the limitations of small, nonrepresentative data sets. This research sought to enhance AI model generalization through the creation of a more diverse, multicenter data set. METHODS: To achieve this goal, the study merged the BabyEar4k public data set with additional images from various centers to create a more comprehensive and varied data set. This data set was used to train and fine-tune CNN models. Performance evaluation was based on established metrics, such as accuracy, sensitivity, and specificity, to assess the models' ability to identify auricular deformities. RESULTS: The fine-tuned CNN models demonstrated strong performance, achieving high scores across various metrics. The integration of data from multiple sources improved the model's generalizability, highlighting the significance of a multicenter approach for AI development in clinical settings. CONCLUSIONS: This study emphasizes the importance of diverse, large-scale data sets in enhancing the performance of AI models for clinical applications. The success of the CNN models in diagnosing ear deformities underscores the potential of AI to revolutionize early diagnosis when backed by comprehensive data sets and standardized diagnostic criteria.