Advanced

On the development of an unsupervised probabilistic algorithm for grayscale fluorescence image segmentation

Brander, Magnus LU (2018) FYTK02 20181
Computational Biology and Biological Physics
Department of Astronomy and Theoretical Physics
Abstract
In the field of computational biology, fluorescence microscopy images often constitute the input source of information. The process of binarization of raw images to delineate interesting objects requires image segmentation into signal and background pixels.

Several methods to perform image segmentation exist, the Otsu method being a popular unsupervised example. The Otsu method's lack of probabilistic predictions in terms of accuracy is a limiting factor when it comes to evaluation of the segmented images and their correctness.

Based on the assumption that the background intensity distribution is Gaussian we present a new unsupervised probabilistic segmentation algorithm. The new algorithm uses Bayesian decision theory to classify... (More)
In the field of computational biology, fluorescence microscopy images often constitute the input source of information. The process of binarization of raw images to delineate interesting objects requires image segmentation into signal and background pixels.

Several methods to perform image segmentation exist, the Otsu method being a popular unsupervised example. The Otsu method's lack of probabilistic predictions in terms of accuracy is a limiting factor when it comes to evaluation of the segmented images and their correctness.

Based on the assumption that the background intensity distribution is Gaussian we present a new unsupervised probabilistic segmentation algorithm. The new algorithm uses Bayesian decision theory to classify pixels as signal and background respectively, and provides a prior estimate for the fractions of correctly classified pixels.

Segmentation tests performed on artificial fluorescent images show that the new algorithm performs significantly better for high level of additive noise than the Otsu method. For a low level of additive noise, the new algorithm performs similarly to the Otsu method. Furthermore, the new algorithm provides a prior estimate for the fraction of correctly classified pixels close to the true values.

We hope the new algorithm will constitute a good alternative to already established methods, offering precise probabilistic image segmentation. (Less)
Popular Abstract
With help of our eyes and brain we as humans excel when it comes to extracting meaningful information from our surrounding. The fact that you can read this text and hopefully understand its message, demonstrates great capabilities in handling complex systems and huge amounts of information.

The invention of the camera enabled decades of attempts and endeavors in perfecting the art of portraying reality to be out-performed; taking a picture can today be done with only a fraction of all the practice that earlier painters needed. When portraying is done by the help of the human hand, mind and sight, there will always be room for subjectivity, and the only certain thing is that two people most likely will portray the same scene differently.... (More)
With help of our eyes and brain we as humans excel when it comes to extracting meaningful information from our surrounding. The fact that you can read this text and hopefully understand its message, demonstrates great capabilities in handling complex systems and huge amounts of information.

The invention of the camera enabled decades of attempts and endeavors in perfecting the art of portraying reality to be out-performed; taking a picture can today be done with only a fraction of all the practice that earlier painters needed. When portraying is done by the help of the human hand, mind and sight, there will always be room for subjectivity, and the only certain thing is that two people most likely will portray the same scene differently. However, with a camera there is no room for statements like "I saw it that way", and humanity can enter a new era of objective book-keeping.

As our access to information through images increase with a modernized society, the era of internet and digitalization, the need of methods to efficiently and accurately extract meaningful information from images increases as well. Image analysis, which aims to extract meaningful information from images, is therefore present in more areas of society that many can imagine. Face recognition features in smartphones, goods inspection in industry and cutting edge research of DNA all rely on image analysis.

In science there is a need of quantifying results and ensuring their reproducibility. Subjective statements are of little value even though they are based on trivial observations. Therefore, one of many challenges in image analysis is to quantify the seemingly obvious.

In the field of computational biology, image analysis is used to extract the information embedded in fluorescence microscopy images. Fluorescent images are for example used to study cancer cells and DNA molecules in closer detail. An important tool in the analysis of fluorescent images consists of image segmentation, which aims to produce a binary image highlighting interesting objects. The work presented in this study approaches the task of achieving optimal binary images, and tries to improve the result as compared to already existing methods designed for this task.

We present a new method to convert fluorescent images into binary images that performs remarkably better than an existing popular method, and in addition to this can predict the accuracy of the achieved extraction of interesting objects. From the predicted accuracy we can, in addition to just reporting a result, also give valuable information to the researcher whether the results can be trusted or not. As knowledge about uncertainty in experiments is of great importance in science we see that the new method could potentially fill that gap found in the methods used today.


We hope that the new method with its benefits for fluorescent image segmentation will help biological and medical research to progress with the image segmentation tasks they are challenged with. (Less)
Please use this url to cite or link to this publication:
author
Brander, Magnus LU
supervisor
organization
course
FYTK02 20181
year
type
M2 - Bachelor Degree
subject
keywords
bayesian decision theory, bayesian, image segmentation, fluorescence microscopy images, Otsu
language
English
id
8948647
date added to LUP
2018-06-14 11:05:52
date last changed
2018-06-14 11:05:52
@misc{8948647,
  abstract     = {In the field of computational biology, fluorescence microscopy images often constitute the input source of information. The process of binarization of raw images to delineate interesting objects requires image segmentation into signal and background pixels. 

Several methods to perform image segmentation exist, the Otsu method being a popular unsupervised example. The Otsu method's lack of probabilistic predictions in terms of accuracy is a limiting factor when it comes to evaluation of the segmented images and their correctness. 

Based on the assumption that the background intensity distribution is Gaussian we present a new unsupervised probabilistic segmentation algorithm. The new algorithm uses Bayesian decision theory to classify pixels as signal and background respectively, and provides a prior estimate for the fractions of correctly classified pixels. 

Segmentation tests performed on artificial fluorescent images show that the new algorithm performs significantly better for high level of additive noise than the Otsu method. For a low level of additive noise, the new algorithm performs similarly to the Otsu method. Furthermore, the new algorithm provides a prior estimate for the fraction of correctly classified pixels close to the true values. 

We hope the new algorithm will constitute a good alternative to already established methods, offering precise probabilistic image segmentation.},
  author       = {Brander, Magnus},
  keyword      = {bayesian decision theory,bayesian,image segmentation,fluorescence microscopy images,Otsu},
  language     = {eng},
  note         = {Student Paper},
  title        = {On the development of an unsupervised probabilistic algorithm for grayscale fluorescence image segmentation},
  year         = {2018},
}