An Innovative 3D Slicer Plugin for Brain Images Annotation and Lesions Study

This article presents the design, development, and validation of a plugin that integrates deep learning models into the open-source 3D Slicer framework for the visualization and analysis of biomedical images. The system implements a convolutional neural network (CNN) based on VGG16, trained on preprocessed medical images (i.e., format conversion, normalization, data augmentation), then converted to ONNX and inferred in Slicer via ONNX Runtime. The plugin enables image loading, inference, and result visualization directly within the Slicer interface, featuring a modular architecture for data acquisition, pre-processing, prediction, and reporting. During validation, the model achieved approximately 85% accuracy and an AUC of 0.94, with better performance on high-definition images.