Abstract
BACKGROUND: The integration of artificial intelligence (AI) in dental implant planning has emerged as a transformative approach to enhance diagnostic accuracy and efficiency. This study aimed to evaluate the performance of two object detection models, Faster R-CNN and YOLOv7 in analyzing cross-sectional and panoramic images derived from DICOM files processed by four distinct dental imaging software platforms. METHODS: The dataset consisted of 332 implant position images derived from DICOM files of 184 CBCT scans. Three hundred images were processed using DentiPlan Pro 3.7 software (NECTEC, NSTDA, Thailand) for the development of Faster R-CNN and YOLOv7 models for dental implant planning. For model testing, 32 additional implant position images, which were not included in the training set, were processed using four different software programs: DentiPlan Pro 3.7, DentiPlan Pro Plus 5.0 (DTP; NECTEC, NSTDA, Thailand), Implastation (ProDigiDent USA, USA), and Romexis 6.0 (Planmeca, Finland). The performance of the models was evaluated using detection rate, accuracy, precision, recall, F1 score, and the Jaccard Index (JI). RESULTS: Faster R-CNN achieved superior accuracy across imaging modalities, while YOLOv7 demonstrated higher detection rates, albeit with lower precision. The impact of image rendering algorithms on model performance underscores the need for standardized preprocessing pipelines. Although Faster R-CNN demonstrated relatively higher performance metrics, statistical analysis revealed no significant differences between the models (p-value > 0.05). CONCLUSIONS: This study emphasizes the potential of AI-driven solutions in dental implant planning and advocates the need for further research in this area. The absence of statistically significant differences between Faster R-CNN and YOLOv7 suggests that both models can be effectively utilized, depending on the specific requirements for accuracy or detection. Furthermore, the variations in imaging rendering algorithms across different software platforms significantly influenced the model outcomes. AI models for DICOM analysis should rely on standardized image rendering to ensure consistent performance.