Visualització de dades mèdiques

Abstract

Visual analysis of medical volume datasets performed by medical experts provides valuable information in research and clinical settings such as radiology or nuclear medicine. Computer visualization methods seek to present to the user an informative display of a given dataset, trying to minimize both the computation time and the user interaction.

The goal of this project is to analyze, design and implement a volume visualitzation system that proposes a new method to improve the visualitzation of segmented volume datasets.

In this scenario, two common volume visualization methods included in commercial medical software are the Direct Volume Rendering (DVR), and the Maximum Intensity Projection (MIP). DVR allows to have a spatial context and MIP is an easy-­to­implement method. Both methods display good visualitzations but have several drawbacks.

This project develops a new method called Maximum Intensity Difference Accumulation (MIDA) that exploits the data characteristics to generate good visualitzation results. The basic idea behind the MIDA approach is combining the advantatges of DVR and MIP and therefore provide an improved visualization framework with reduced limitations. Also, in this project an adaptation of the MIDA framework is proposed to visualize the volume in conjuntion with any segmentation mask of it.

MIDA is a general method that can be applied to several types of volume datasets. In particular, this project focuses on the application of the proposed method in Positron Emission Tomography (PET) volumes obtained from the Nuclear Medicine Department at the H. Sant Pau. PET is a commonly used nuclear medicine diagnosis test to detect and quantify the presence of any tumor volume within a patient’s body, where physicians may also provide tumor segmentation masks.

Visualization of segmented PET scans using our proposed MIDA­based system provided superior visual results with respect to DVR and MIDA approaches. A GPU implementation provides substancial runtime improvement with respect to a CPU approach, obtaining the real­time visualization needed in the clinical setting.