Fast Maximum Intensity Projections of Large Medical Data Sets by Exploiting Hierarchical Memory Architectures

3D Vascular Segmentation Supervised by 2D Annotation of Maximum Intensity Projection

Vascular structure segmentation plays a crucial role in medical analysis and clinical applications. The practical adoption of fully supervised segmentation models is impeded by the intricacy and time-consuming nature of annotating vessels in the 3D space. This has spurred the exploration of weakly-supervised approaches that reduce reliance on expensive segmentation annotations. Despite this, existing weakly supervised methods employed in organ segmentation, which encompass points, bounding boxes, or graffiti, have exhibited suboptimal performance when handling sparse vascular structure. To alleviate this issue, we employ maximum intensity projection (MIP) to decrease the dimensionality of 3D volume to 2D image for efficient annotation, and the 2D labels are utilized to provide guidance and oversight for training 3D vessel segmentation model. Initially, we generate pseudo-labels for 3D blood vessels using the annotations of 2D projections. Subsequently, taking into account the acquisition method of the 2D labels, we introduce a weakly-supervised network that fuses 2D-3D deep features via MIP to further improve segmentation performance. Furthermore, we integrate confidence learning and uncertainty estimation to refine the generated pseudo-labels, followed by fine-tuning the segmentation network. Our method is validated on five datasets (including cerebral vessel, aorta and coronary artery), demonstrating highly competitive performance in segmenting vessels and the potential to significantly reduce the time and effort required for vessel annotation. Our code is available at: https://github.com/gzq17/Weakly-Supervised-by-MIP.

Interactive GPU-based maximum intensity projection of large medical data sets using visibility culling based on the initial occluder and the visible block classification

Maximum intensity projection (MIP) is an important visualization method that has been widely used for the diagnosis of enhanced vessels or bones by rotating or zooming MIP images. With the rapid spread of multidetector-row computed tomography (MDCT) scanners, MDCT scans of a patient generate a large data set. However, previous acceleration methods for MIP rendering of such a data set failed to generate MIP images at interactive rates. In this paper, we propose novel culling methods in both object and image space for interactive MIP rendering of large medical data sets. In object space, for the visibility test of a block, we propose the initial occluder resulting from a preceding image to utilize temporal coherence and increase the block culling ratio a lot. In addition, we propose the hole filling method using the mesh generation and rendering to improve the culling performance during the generation of the initial occluder. In image space, we find out that there is a trade-off between the block culling ratio in object space and the culling efficiency in image space. In this paper, we classify the visible blocks into two types by their visibility. And we propose a balanced culling method by applying a different culling algorithm in image space for each type to utilize the trade-off and improve the rendering speed. Experimental results on twenty CT data sets showed that our method achieved 3.85 times speed up in average without any loss of image quality comparing with conventional bricking method. Using our visibility culling method, we achieved interactive GPU-based MIP rendering of large medical data sets.