Image Analysis and Machine Learning Techniques for Digital Histopathological Images

• Author / Creator

  Salah Mohammed Awad AL-Heejawi

• The histopathological examination of tissue biopsies is used as the gold standard by pathologists for diagnosing skin cancers, such as melanoma. Traditionally, the histopathological slides are examined by clinicians under a microscope. With recent advances in digital imaging and scanning, a glass slide biopsy can now be digitized at high magnifications to create a high resolution digital whole slide image (WSI), which facilitates pathologists for disease identification by analyzing a WSI image on a computer monitor. However, the amount of data produced by a whole slide digital scanner, which can exceed billions of pixels for a WSI, manual analysis of an image can take significant amount of time. More importantly, the manual analysis by clinicians is typically subjective and often prone to intra- and inter-observer variability. The main goal of this dissertation is to develop image analysis and machine learning algorithms that can aid towards building a Computer-Aided Image Analysis (CAIA) system for digital skin histopathological images. These systems would automatically extract meaningful features from the WSI and perform classification and grading to help pathologists to obtain prompt and accurate diagnosis.
  In this dissertation, we propose four CAIA systems to grade and detect melanoma based on histopathological image analysis. The proliferative index is a useful indicator for cancer grading. First, a technique is proposed for efficient segmentation of the lymph nodes in a lymph node WSI (obtained using stains such as H&E, MART-1, and Ki-67) and estimates the rate of cell growth by detecting the actively proliferative nuclei using classical machine learning techniques. Secondly, a novel Convolutional Neural Network (CNN) architecture is developed for efficient segmentation of nuclei regions and a seed detection technique to accurately detect the number of nuclei. These two techniques together can be used to obtain an accurate estimation of the proliferation index. Thirdly, a CAIA system is proposed where the nuclei are segmented using a CNN architecture, and a hand-crafted feature-based technique is used to detect the melanoma nuclei on H&E-stained lymph node WSIs. Finally, a CNN-based technique is proposed for automated detection of the melanoma regions on H&E-stained skin histopathological images. Experimental results show significant performance improvement in the detection and grading of melanoma over state-of-the- art techniques.

• Subjects / Keywords

  • Machine learning
  • image processing
  • histopathological images
  • melanoma cancer

• Graduation date

  Fall 2021

• Type of Item

  Thesis

• Degree

  Doctor of Philosophy

• DOI

  https://doi.org/10.7939/r3-6baa-aq58

• License

  This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.