[Paper Review] Unsupervised Segmentation Algorithms' Implementation in ITK for Tissue Classification via Human Head MRI Scans

Shadman Sakib, Md. Abu Bakr Siddique|arXiv (Cornell University)|Feb 26, 2019

Medical Image Segmentation Techniques14 references4 citations

TL;DR

This paper implements and evaluates three unsupervised segmentation algorithms—Otsu thresholding, Bayesian classification, and Bayesian classification with Gaussian smoothing—within the Insight Toolkit (ITK) for tissue classification in human head MRI scans. Using five T1-weighted MRI datasets, the study demonstrates that Bayesian classification with Gaussian smoothing achieves the most accurate segmentation of white matter, gray matter, CSF, and background tissues, outperforming the other methods in consistency and anatomical fidelity.

ABSTRACT

Tissue classification is one of the significant tasks in the field of biomedical image analysis. Magnetic Resonance Imaging (MRI) is of great importance in tissue classification especially in the areas of brain tissue classification which is able to recognize anatomical areas of interest such as surgical planning, monitoring therapy, clinical drug trials, image registration, stereotactic neurosurgery, radiotherapy etc. The task of this paper is to implement different unsupervised classification algorithms in ITK and perform tissue classification (white matter, gray matter, cerebrospinal fluid (CSF) and background of the human brain). For this purpose, 5 grayscale head MRI scans are provided. In order of classifying brain tissues, three algorithms are used. These are: Otsu thresholding, Bayesian classification and Bayesian classification with Gaussian smoothing. The obtained classification results are analyzed in the results and discussion section.

Motivation & Objective

• To develop and integrate unsupervised segmentation algorithms into the Insight Toolkit (ITK) for biomedical image analysis.
• To classify four brain tissue types—white matter, gray matter, cerebrospinal fluid (CSF), and background—using unsupervised learning on T1-weighted MRI scans.
• To compare the performance of Otsu thresholding, standard Bayesian classification, and Bayesian classification with Gaussian smoothing in segmenting human brain tissues.
• To assess the robustness and accuracy of these methods on real clinical MRI data without ground truth labels.

Proposed method

• The study uses five T1-weighted human head MRI scans as input for unsupervised tissue classification.
• Otsu thresholding is applied to automatically determine optimal intensity thresholds for tissue separation based on image histogram distribution.
• Bayesian classification is implemented using class-conditional probability density functions estimated from image intensity statistics.
• Gaussian smoothing is applied to the input MRI data before Bayesian classification to reduce noise and improve class separability.
• Segmentation results are evaluated qualitatively through visual inspection and quantitative analysis of tissue distribution and spatial consistency.
• All algorithms are implemented in ITK, leveraging its C++ framework for efficient and reproducible biomedical image processing.

Experimental results

Research questions

• RQ1 How accurately can unsupervised segmentation algorithms classify white matter, gray matter, CSF, and background in human head MRI scans?
• RQ2 What is the impact of Gaussian smoothing on the performance of Bayesian classification in tissue segmentation?
• RQ3 How do Otsu thresholding, standard Bayesian classification, and smoothed Bayesian classification compare in terms of anatomical accuracy and noise robustness?
• RQ4 Can unsupervised methods in ITK produce reliable and reproducible tissue classification results without ground truth labels?

Key findings

• Bayesian classification with Gaussian smoothing produced the most anatomically plausible segmentation results, with clearer boundaries between white matter, gray matter, and CSF.
• Otsu thresholding showed limited effectiveness due to overlapping intensity distributions among tissue classes, leading to fragmented and inaccurate segmentations.
• Standard Bayesian classification performed better than Otsu but still exhibited sensitivity to noise and intensity inhomogeneities in the MRI scans.
• The addition of Gaussian smoothing significantly improved the consistency and spatial coherence of tissue classification, especially in regions with low contrast or partial volume effects.