YOLO_MIC3: Facial Keypoints Detection Algorithm for Aggregating Reduced-Dimensional MRI Images

Clinical diagnosis and treatment are highly dependent on medical images. However, the MRI reconstruction model of the human head is limited by insufficient resolution, blurred tissue contours, and spatial impurities, which makes it difficult to accurately locate the key points of the face, and thus severely restricts its wide application in clinical scenarios. In this study, we take a unique approach to capture the MRI reconstruction model with dimensionality reduction, and construct the YOLO_MIC3 deep learning algorithm based on YOLOv8 for the obtained 2D images. Through a series of optimization strategies, such as introducing the CBAM mechanism, replacing the DarknetBottleneck component in C2f with the CAA module, and applying the Inner-CloU computational loss, the comprehensive performance of the model is improved. Experimentally, the algorithm reduces the model complexity by 25%, improves the computational efficiency by 29%, and performs well in the Box metrics dimension of mAP50-95, which provides a promising solution for the localization of clinical facial keypoints.