Resumen de Real-time quality visualization of medical models on comodity and mobile devices

Jesús Díaz García

• This thesis concerns the specific field of visualization of medical models using commodity and mobile devices.

  Mechanisms for medical imaging acquisition such as MRI, CT, and micro-CT scanners are continuously evolving, up to the point of obtaining volume datasets of large resolutions (> 512^3). As these datasets grow in resolution, its treatment and visualization become more and more expensive due to their computational requirements. For this reason, special techniques such as data pre-processing (filtering, construction of multi-resolution structures, etc.) and sophisticated algorithms have to be introduced in different points of the visualization pipeline to achieve the best visual quality without compromising performance times. The problem of managing big datasets comes from the fact that we have limited computational resources. Not long ago, the only physicians that were rendering volumes were radiologists. Nowadays, the outcome of diagnosis is the data itself, and medical doctors need to render them in commodity PCs (even patients may want to render the data, and the DVDs are commonly accompanied with a DICOM viewer software). Furthermore, with the increasing use of technology in daily clinical tasks, small devices such as mobile phones and tablets can fit the needs of medical doctors in some specific areas. Visualizing diagnosis images of patients becomes more challenging when it comes to using these devices instead of desktop computers, as they generally have more restrictive hardware specifications.

  The goal of this Ph.D. thesis is the real-time, quality visualization of medium to large medical volume datasets (resolutions >= 512^3 voxels) on mobile phones and commodity devices.

  To address this problem, we use multiresolution techniques that apply downsampling techniques on the full resolution datasets to produce coarser representations which are easier to handle.

  We have focused our efforts on the application of Volume Visualization in the clinical practice, so we have a particular interest in creating solutions that require short pre-processing times that quickly provide the specialists with the data outcome, maximize the preservation of features and the visual quality of the final images, achieve high frame rates that allow interactive visualizations, and make efficient use of the computational resources.

  The contributions achieved during this thesis comprise improvements in several stages of the visualization pipeline.

  The techniques we propose are located in the stages of multi-resolution generation, transfer function design and the GPU ray casting algorithm itself. More precisely, the contributions achieved during this thesis are: * A feature-preserving filter for the downsampling of scalar volume data.

  * A downsampling filter and an efficient storage scheme for gradient volume data.

  * An algorithm for the modification of transfer functions to improve the visualization quality of coarse levels.

  * A multiresolution framework and two progressive raycasting strategies for the incremental rendering of volume data.