Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
A CAD system was developed to detect caries and molar-incisor hypomineralization (MIH) in intraoral photographs using a modified YOLOv5 architecture called Intra-YOLO. Transfer learning was employed to leverage pre-trained weights, and reinforcement learning was used to optimize the model's hyperparameters for improved detection accuracy of small, visually similar lesions. The resulting Intra-YOLO model demonstrated improved performance in detecting caries and MIH compared to standard YOLOv5, addressing the challenges posed by small object detection and lesion similarity.
Reinforcement learning can fine-tune a YOLOv5 model to drastically improve detection of visually similar, small dental lesions in intraoral photography, outperforming standard transfer learning.
This study developed a computer-aided diagnosis (CAD) system for detecting caries and molar-incisor hypomineralization (MIH) in intraoral photographs. These lesions share similar appearances, making clinical differentiation challenging, especially given their small size and variability in imaging conditions.