Published on May 17th, 2013 | by Emily Corbett0
Predicting coronary heart disease
By Tanzila Chowdhury, The University of Western Sydney
Coronary heart disease is the leading cause of death in most countries. Early detection can save thousands of lives. The first step for early detection is getting a clear picture of movement of the blood vessels within the heart. Angiograms allow us to look inside the coronary arteries, helping us to determine the severity of narrowing in blood vessels. Unfortunately, angiograms give two dimensional (2D) images which can provide misleading information. Three dimensional (3D) images provide us with exact representation. Hence, creating 3D models from 2D images can be very beneficial.
2D angiogram images of a patient’s heart are collected from camera viewpoints at two different angles. Images are collected when the heart is full of blood (end-diastolic phase) and as empty as possible (end-systolic phase). Our main goal for this project was to reconstruct the 3D model of a heart using these two pairs of images. We examined two different methods: the eight point algorithm and five point algorithm, and examined which gave a better 3D representation of our data. The steps required for the 3D reconstruction process are:
1. Calculate the fundamental matrix and the essential matrix.
2. Calculate the camera matrices.
3. Calculate the point in space that projects to the image point correspondence.
The first aim for this project was to compare these different methods to find which method gives a more accurate 3D model. We evaluated both the methods and projected back the 3D points to compare with the actual points. Although we get very accurate results with the 5 point method for one camera image, the 8 point method gives overall better results.
During our analysis we found that even though the raw data points provided by the specialist were chosen carefully, there was error in their positioning. Therefore, the second aim was to examine the effect of error when performing the 3D reconstruction. When comparing the results of the 8 point and 5 point algorithms and the effect of the error, we found that the 8 point method provided better preservation for the shape of the arteries, which is important for later analysis. From these results, we conclude that, the 8 point method is more robust to the errors than 5 point method. Hence, the 8 point method is more suitable to reconstruct a 3D model of our data.
This student took part in the 2012/13 AMSI Vacation Research Scholarship program. For more information on this years program please click here