Diabetes is a metabolic disease where the insulin system of a patient is not working properly. In 2013, about 380 million people worldwide were estimated to have diabetes, causing about 5.1 million deaths every year. Despite these large numbers, patients are frequently diagnosed after onset of symptoms. Developing techniques to obtain information about the biochemical properties of the pancreas in situ would improve understanding of diabetes development. Applying such techniques in vivo would dramatically increase the possibilities to diagnose and characterize diabetes. This work is part of a project where the potential of vibrational spectroscopy for the diagnosis of diabetes is assayed.
For this thesis, data of the same sample (a bright-field image and a matrix of spectral data) from two different instrument types is used. Due to handling issues, the data sets do not represent the exact same areas of the sample. My task was therefore to develop a program in MATLAB that identifies the overlapping areas based on the bright-field images and crops the corresponding spectral data to the correct size.
The images available during development of this program did not contain enough information usable for an algorithm to detect identical positions to calculate an overlay, since the images were in black and white and only displayed arbitrary lines. Additionally, the images were scaled (both along the X- and the Y-axis) and rotated towards each other, increasing the complexity of the task. Therefore, user input was necessary to obtain an overlay of the images from both data sources.
In biomedical imaging, the wider surroundings of a pixel are still important to obtain as much information from the data as possible. In this work, a group of image pixels correspond to a data point in a matrix of spectral absorptions. These two constraints resulted in the development of a function that identified the largest rectangle with approximately equal length and height in the overlapping area.
2015. , 35 p.