Tomoya Kitano

Background/Objectives: Vision-language models (VLMs) such as GPT (Generative Pre-Trained Transformer) have shown potential for medical image interpretation; however, challenges remain in generating reliable radiological findings in clinical practice, as exemplified by dental pathologies. This study proposes a Self-correction Loop with Structured Output (SLSO) framework as an integrated processing methodology to enhance the accuracy and reliability of AI-generated findings for jaw cysts in dental panoramic radiographs. Methods: Dental panoramic radiographs with jaw cysts were used to implement a 10-step integrated processing framework incorporating image analysis, structured data generation, tooth number extraction, consistency checking, and iterative regeneration. The framework functioned as an external validation mechanism for GPT outputs. Performance was compared against the conventional Chain-of-Thought (CoT) method across seven evaluation items: transparency, internal structure, borders, root resorption, tooth displacement, relationships with other structures, and tooth number. Results: The SLSO framework improved output accuracy for multiple items compared to the CoT method, with the most notable improvements observed in tooth number identification, tooth displacement detection, and root resorption assessment. In successful cases, consistently structured outputs were achieved after up to five regenerations. The framework enforced explicit negative finding descriptions and suppressed hallucinations, although accurate identification of extensive lesions spanning multiple teeth remained limited. Conclusions: This investigation established the feasibility of the proposed integrated processing methodology and provided a foundation for future validation studies with larger, more diverse datasets.

Purpose: This study investigated whether or not cervical vertebral morphology could be automatically classified using a deep-learning system.Methods: Nine hundred and ninety-nine cephalometric radiographs obtained for orthodontic diagnoses were used in the present study. Segmented rectangular images between the second and fourth cervical vertebrae were resized with a patch size of 300 × 300 pixels (resized images) or padded with black images on the right and left sides, keeping their aspect ratio unchanged (padded images). Seven hundred and ninety-nine and 200 images were defined as the training and test data, respectively. Le Net and Alex Net Neural Network Console architectures were used to classify the images into two groups based on cervical vertebral maturation (CVM). The classifications of resized images using Le Net and Alex Net were designated Le Net Resizing and Alex Net Resizing, respectively. The classifications of the padded images using Le Net and Alex Net were designated Le Net Padding and Alex Net Padding, respectively. The accuracy of the classification using deep learning was evaluated.Results: The average accuracies of the classification of CVM using Le Net Resizing and Le Net Padding were 59.0% and 89.5%, respectively. The average accuracy of classification of CVM using Alex Net Resizing and Alex Net Padding was 90.5% and 93.0%, respectively.Conclusion: The deep-learning system using Le Net Padding, Alex Net Resizing, and Alex Net Padding showed high accuracy in classifying images into two groups based on CVM.

Objective This study explored the feasibility of using deep learning for profiling of panoramic radiographs. Study design Panoramic radiographs of 1000 patients were used. Patients were categorized using seven dental or physical characteristics: age, gender, mixed or permanent dentition, number of presenting teeth, impacted wisdom tooth status, implant status, and prosthetic treatment status. A Neural Network Console (Sony Network Communications Inc., Tokyo, Japan) deep learning system and the VGG-Net deep convolutional neural network were used for classification. Results Dentition and prosthetic treatment status exhibited classification accuracies of 93.5% and 90.5%, respectively. Tooth number and implant status both exhibited 89.5% classification accuracy; impacted wisdom tooth status exhibited 69.0% classification accuracy. Age and gender exhibited classification accuracies of 56.0% and 75.5%, respectively. Conclusion Our proposed preliminary profiling method may be useful for preliminary interpretation of panoramic images and preprocessing before the application of additional artificial intelligence techniques.