Department of Computer Science and Engineering, University of Minnesota, Minneapolis, MN
Department of Laboratory Medicine and Pathology, University of Minnesota Medical Center, Fairview, Minneapolis, MN
Early diagnosis is quintessential for the effective treatment of disease. For a surgical pathologist, the most time-consuming aspect of the diagnostic process involves arduously scrutinizing tissue slides under a microscope for evidence of disease. As a result, even a skilled pathologist can only diagnose a few patients per day. Our goal is to expedite this process through computer-assisted diagnosis. As a significant step in this direction, our goal is to develop novel image processing and machine learning algorithms tailored toward automated and semi-automated cancer diagnosis. Preliminary results on endometrioid and adenocarcinoma were favorable, with overwhelming positive feedback from the community.
Specifically, the goal of this project is to reduce the time spent by pathologists on each tissue slide. To achieve this, we focus on three main sub-tasks. The first step is automatic segmentation, in which we aim to direct the attention of the pathologist towards the regions which are likely to be cancerous with a minimal false negative (miss) and false positive (false alarm) rate. Typical outputs of this step are `high confidence cancer regions', 'low confidence cancer regions', and 'high confidence benign regions', determined automatically by the segmentation step.
Based on the segmented regions, our second step is to identify the disease associated with the candidate malign regions. This process entails building appearance-based modeling for capturing the unique visual cues that a pathologist uses to identify a disease. There is a significant contribution of low-level image features and high-level computer vision/ machine learning algorithms.
As a final step, we plan on using the data-driven computational models of the different pathological conditions to interpret and confirm more complicated medical conditions. Our specific objective is for experts to use this framework as a tool for discovering and developing a deeper insight into these medical conditions. For example, we are interested in determining the "fine-line" or boundaries between the different stages of cancer.
Projects
Diagnosing Endometrial Carcinoma via Computer-Assisted Image Analysis | 
Diagnosing Adenocarcinoma of the Prostate by Computer Vision Methods |
Publications
- Diagnosing Adenocarcinoma of the Prostate by Computer Vision Methods
Ravishankar Sivalingam, Guruprasad Somasundaram, Xinyan Li, Alesia Kaplan, Jonathan Henriksen, Arindam Banerjee, Vassilios Morellas, Nikolaos Papanikolopoulos, and Alexander Truskinovsky
Annual Meeting of the United States & Canadian Academy of Pathology (USCAP), 2012 - Diagnosing Endometrial Carcinoma via Computer-Assisted Image Analysis
Ravishankar Sivalingam, Guruprasad Somasundaram, Aravind Ragipindi, Arindam Banerjee, Vassilios Morellas, Nikolaos Papanikolopoulos, and Alexander Truskinovsky
Annual Meeting of the United States & Canadian Academy of Pathology (USCAP), 2011.
People
Faculty
Dr. Nikolaos Papanikolopoulos
Dr. Vassilios Morellas
Dr. Arindam Banerjee
Dr. Alexander Truskinovsky, M.D.
Dr. Alesia Kaplan, M.D.
Graduate Students
Ravishankar Sivalingam, Ph.D. (graduated)
Guruprasad Somasundaram, Ph.D. (graduated)
Xinyan Li, Ph.D. (graduated 2021)
Alumnus
Aravid Ragipinda
This work is supported by grants from the University of Minnesota Institute for Engineering in Medicine and the National Science Foundation. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.