Intergral Cardiovascular Image Interpretation

Towards Automated Cardiovascular Image Interpretation: an Integral Approach

Dutch Science Foundation NWO, Innovational Research Incentive, "vernieuwingsimpuls", awarded 2001
Boudewijn Lelieveldt, Ph.D, Avan Suinesiaputra, M.Sc

Background

Cardiovascular disease poses a major health problem in the western world, and early and accurate diagnosis is critically important for successful treatment. Recently, modern imaging techniques such as Magnetic Resonance Imaging (MRI) and Multi-Slice Computed Tomography (MSCT) have become recognized as powerful diagnostic modalities, enabling such a detailed assessment of the presence and extent of cardiovascular disease. Both modalities are characterized by their 4D nature of the image data and high spatial and temporal resolutions, with MRI having the advantage of absence of ionizing radiation. Latest developments also indicate that the coronary arterial wall and plaque composition can be determined by MRI. MSCT at present provides a higher resolution of the coronary arteries, and is the only technique that allows the assessment of calcium in the coronaries. Apart from diagnostic applications, MRI and MSCT also play an important role in understanding the mechanisms involved in the progression of cardiovascular disease.
Because of the rapid progress in cardiovascular MR and CT acquisition techniques, a new problem has surfaced: the amount of image data involved in a comprehensive patient study is massive (1500-5000 images.) This makes it a formidable task for the radiologist, cardiologist or technician to interpret the image data, let alone analyze all these images quantitatively with simple measurement tools. Moreover, manual analysis is subjective and compromises the accuracy and reproducibility of quantitative measurements. These factors have generated a great demand for tools that facilitate and further automate the accurate analysis of cardiac MR and CT patient examinations. In other words, what the radiologist / cardiologist needs is a “single analysis button” that executes the automated extraction of clinically relevant information about the function of the heart and the status of the coronary arteries from these thousands of images, all within a matter of minutes.

Goals

The goal of this research program is the development of novel, automated and highly robust analysis tools to facilitate the interpretation of cardiac MR and MSCT patient examinations in routine clinical practice and clinical research with minimal human interaction. Ideally, a novel system is envisioned that will automatically recognize and present the radiologist /cardiologist with diagnostic evidence for a suspected disease based on the imaging parameters. Currently, no such tools exist. To realize this goal, we focus on the development of novel computer algorithms for:

  • On-line analysis of global and regional left- and right ventricular function in (real-time) cardiac MR images,
  • On-line analysis of ventricular wall motion from MR tagging images and velocity encoded MR images,
  • Time-continuous, trainable and adaptive contour detection that also performs robustly in exceptional cases and images of deformed patient hearts, and that is able to "learn" from new cases and observer preferences,
  • Consistent and reproducible contour detection in stress-acquisitions and follow-up studies,
  • Reliable functional and anatomical quantification of coronary narrowings from MSCT and MR images,
  • Automated quantification of regional myocardial perfusion,
  • Computer-assisted diagnosis.

These components all contribute complementary pieces to the diagnostic puzzle: the key novelty of this project lies in combining these pieces in an analysis tool for an integral patient examination. As a result, a far more complete diagnostic picture emerges than would be possible with conventional approaches, which focus on single image sets.

Approach

To realize this goal, we will tackle an array of technical challenges, with the most important five being:

Mathematical representation of a-priori knowledge. Robust automated analysis algorithms will heavily rely on expert knowledge about cardiac shape, motion, appearance, pathology and physiology. One of the key challenges of this proposal is the development and extensive evaluation of techniques that integrate expert anatomical knowledge into the automated analysis of cardiovascular images,

Integral fusion of all available image data. In cardiovascular MR and MSCT imaging protocols, multiple image sets are acquired covering the same anatomical area. Conventional post-processing methods analyze one image set at a time. In this project however, we want to prove, that by handling post-processing in the context of an integral patient examination and by simultaneously fusing information from all available image sets, we greatly increase the robustness and reliability of automated contour detection methods,

Development of generic methods to handle multiple, potentially conflicting interpretations and to deal with uncertainty in the contour detection by designing segmentation methods in a multi-agent architecture,

Development of computer-assisted diagnostic methods. We will develop separation techniques for disease-specific discriminant features that can be applied to recognize and highlight suspicious areas in the data,

Evaluation of the novel technologies in a clinical setting in collaboration with several clinical partners.

Status

The project started on 1 januari 2002, and is currently in the startup phase.

Contact

For further information, please contact:
Boudewijn P.F. Lelieveldt, Ph.D
Division of Image Processing
Department of Radiology, 1-C2S
Leiden University Medical Center
P.O. Box 9600
2300 RC Leiden
The Netherlands
Tel. +31 (0)71 526 4846
Fax. +31 (0)71 526 6801
e-mail: B.P.F.Lelieveldt@lumc.nl