automatic classification of histopathological images

Open Access
- Author:
- Gillespie, James Joseph
- Area of Honors:
- Electrical Engineering
- Degree:
- Bachelor of Science
- Document Type:
- Thesis
- Thesis Supervisors:
- Vishal Monga, Thesis Supervisor
Dr. Julio V. Urbina, Thesis Honors Advisor - Keywords:
- image classification
histopathological
image processing - Abstract:
- Automated image classification has become extremely desirable in the medical field for faster and more accurate diagnosis of disease. These systems have also gained significant attention because they help doctors in evaluating large volumes of medical imagery. An automated classification system has been developed that categorizes histopathological images provided by the Animal Diagnostic Laboratory (ADL) at the Pennsylvania State University. The classification system has been developed for four different mammalian organs: liver, lung, kidney, and spleen. The classifier automatically labels each histopathological image as healthy, inflammatory, or necrotic (liver only). This system uses supervised classification, which consists of two key stages. First, there is a feature extraction stage where discriminative image features are obtained. Next, these features are fed into a decision engine that performs the class assignment. The primary focus of the research proposed in this paper is the development of a robust feature extraction algorithm for histopathological image classification. Because of its widespread use as a state of the art classifier, Support Vector Machines are used as the decision engine in this research. After training the system with 20 images per class per organ, the system is tested using 30 images per class per organ, and produces classification accuracy up to 97%. These results prove the validity in continuing to improve the feature extraction and investigate different classifiers to provide an even more robust automated classification system for histopathological images.