Scaling Self-Supervised and Cross-Modal Pretraining for Volumetric CT Transformers

A new framework named ReBrain has been introduced for reconstructing brain MRI from sparse CT slices using a retrieval-augmented diffusion approach. This method utilizes a Brownian Bridge Diffusion Model to synthesize MRI slices and retrieves similar CT slices from a prior database to enhance reconstruction accuracy.

A new statistical framework for crack pre-localization in 3D concrete images has been introduced, addressing the challenges of effectively segmenting cracks in large-scale computed tomography (CT) images. This method utilizes a Hessian-based filter and geometric descriptors to identify regions likely to contain cracks, relying on minimal calibration data rather than extensive annotated datasets.

A new publicly accessible model for automated segmentation of muscle and fat in computed tomography (CT) images has been introduced, enhancing body composition analysis. This model effectively segments skeletal muscle, subcutaneous adipose tissue, and visceral adipose tissue in axial CT images, addressing a significant gap in available tools for clinical applications.

The introduction of CleverDistiller marks a significant advancement in self-supervised cross-modal knowledge distillation, enabling the transfer of features from 2D vision foundation models to 3D LiDAR-based models. This framework utilizes a direct feature similarity loss and a multi-layer perceptron projection head, enhancing the learning of complex semantic dependencies in autonomous driving applications.