Loading...

Cancer Detection and Classification in Histopathology Images Under Small Training Set

Askari Farsangi, Amir Hossein | 2022

814 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55133 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Rohban, Mohammad Hossein; Sharifi Zarchi, Ali
  7. Abstract:
  8. Histopathology images are a type of medical images that are used to diagnose a variety of diseases. One of these illnesses is the Leukemia cancer, which has four different subtypes and is diagnosed using a blood smear image. As a result of the advancement of deep learning tools, models for diagnosing various types of disease from images have been developed in recent years.In this project, one of the best models developed to diagnose four different types of disease was replicated, and it was demonstrated that, while this model achieves acceptable accuracy, its decision is not based on medically significant criteria. In the following, a general method for diagnosing the disease is proposed based on some important medical criteria. This method is comprised of the following steps: detection of white blood cells; analysis of each cell image; summarization of the results; and final judgment. Cell detection in blood smear images, a white blood cell classifier, and a segmentation tool are required to develop this method. Each of these tools has been thoroughly investigated in this project, and new and significant achievements have been made in relation to them, such as a robust method for white blood cell classification and high-quality cell segmentation with a small dataset. It was demonstrated that there is an inherent challenge in decision-making based on images for the final diagnosis of Leukemia in the case of simultaneous diagnosis of all types of disease and that to diagnose, a large number of images of a patient must be examined. For this purpose, a model has been developed that overcomes this problem as much as possible and can achieve an accuracy of 81.79 percent in the simplified problem of distinguishing between a healthy sample, a chronic patient, and an acute patient. Finally, an interpretable LSTM-based model was developed to distinguish between ALL and healthy samples, which can achieve 91.11 accuracy in diagnosis while overcoming this inherent challenge. Because of the limited data sets, the models are forced to use a small training data set at all stages of this project
  9. Keywords:
  10. Cancer Diagnosis ; Data Augmentation ; Small Training Set ; Histopathology Images ; Leukemia ; Image Processing

 Digital Object List

 Bookmark

...see more