Honours Projects 2008

Projects supervised by Jinman Kim, Tom Cai and David Feng

Interaction in Biomedical Image Analysis and Visualization
With the introduction of the next-generation of multi-modality medical imaging scanners, new diagnostic capabilities introduced are resulting in improved health care. However, these modern scanners are restricted from its full usage capacity due to the lack of interactive visualization and analysis for diagnosis, where slice-by-slice approach is currently the norm. The massive number of images (1000’s) and the complex inter-relations between the functional (PET) and anatomical (CT) images will mean that access to and assimilation of critical data within these images, by the reader and end-user (e.g. neurosurgeon, cardiothoracic surgeon) will become a major problem. Thus a new method of visualization and interaction is essential. In this project, we will develop a new data fusion algorithm in order to maximize the visual information in an interactive volume rendering of dual-modal PET/CT images. Image fusion is the process of creating new information from combining multiple image sources (data reduction) for optimal visualization. This interactive fusion approach will be based on an iterative approach based on an information energy gain/loss function. This project will build up on a world-class medical visualization system developed in-house, and present the possibility to work with clinical staffs at the Royal Prince Alfred Hospital.

Integrating Functional and Anatomical Images for Simultaneous Tumor Segmentation and Localization
The rapid growths in biomedical image dimensions, in parallel with information technologies, are constantly introducing brand new capabilities and improvement in healthcare, i.e., virtual colonoscopy and 3D visualization. However, access to and assimilation of critical data within these complex images, by the end-user (e.g. neurosurgeon, cardiothoracic surgeon) is becoming a major issue. Image segmentation, a key component in computer aided diagnosis (CAD), enables interactive selection of e.g. tumor regions, and in the process, can dramatically improve image diagnosis. This project will investigate an automated segmentation of tumor regions from dual-modal PET/CT data by integrating functional and anatomical images. The anatomical information from CT will be used to map the PET/CT into a global atlas (registration), providing localization of the tumor regions. As a second stage, deformable model segmentation will be used to grow the tumor regions. This project will present the possibility to work with staffs at the Royal Prince Alfred Hospital.

Spatial Relationships in Biomedical Images for Content-based Retrieval
With the ever-growing image databases in the hospital and other clinical environments, the keyword-based searching and storage solutions currently employed are restricting the full accessibility of the information available in these images. There are increasing demands in image retrieval capabilities, such as the need to retrieve spatially-relevant (visually) images for cancer patients, i.e., evidence-based diagnosis, medical education, and biomedical research. This study will develop a novel method to enable clinical staff to exploit the knowledge of the relationships they know about anatomical/functional structures within the images, to search for similar images. Spatial features from co-registered functional and anatomical PET/CT images will be segmented and the relationships between these features will be used to construct invariant graph representations. These graphs will then be used to search for relevant images that share relevant spatial relationships among the features. This project will extend on already established content-based retrieval system developed in-house, and present the possibility to work with staffs at the Royal Prince Alfred Hospital.

Multi-Modality Medical Data Representation for Mobile Devices
Mobile devices (such as Smartphones, PDAs and UMPCs) have a great many limitations that researchers must consider when developing visualization applications. In particular, displays are limited in size and colour depth, the width/height ratio differs greatly from standard computing devices, onboard hardware is much less powerful and input techniques are very different. These limitations are largely a consequence of the physical size and the usage paradigm of the device. As such, the adaptation of medical data for mobile visualization presents a unique set of challenges for biomedical researchers. Image-processing algorithms can be used to abstract the data, provide overview visualization of medical scans, and reduce the often dizzying number of scroll and zoom operations typically required for comprehensively analyzing clinical data on mobile devices. This research will involve trialing visualization methods such as region-of-interest/context-driven visualization and identifying representation methods which produce worthwhile information enabling improved mobile display and navigation of multi-modality datasets, and building an interface that uses this information to provide visual references that augment the limited mobile display. This project will build up on a world-class medical visualization system developed in-house, and present the possibility to work with clinical staffs at the Royal Prince Alfred Hospital.