Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


thesis_final.pdf (1.46 MB)

ETD Abstract Container

Abstract Header

Machine Learning Models of Histopathologic Images to Serve as a Proxy to Predict Recurrence in ER+/HER- Breast Cancers

Vroom, Carolyn Marie
http://rave.ohiolink.edu/etdc/view?acc_num=osu1661139111525638

Abstract Details

2022, Master of Science, Ohio State University, Computer Science and Engineering.
Digital Pathology lies at the critical intersection of technology and medicine. Leveraging powerful scanners to capture minute details of tissue biopsies by encapsulating multiple magnifications enables a considerable depth of patient information to be consolidated. This avenue of information is a recent development as whole slide scanners have only been widely used in the last ten years after being developed in the 1990s [37]. Thus, exploring whole slide images as another source of patient data has great potential for the medical field for several complex tasks in diagnosis, prognosis, prediction, and classification, especially when there exists a myriad of pressure points when using other patient data methods, such as genetic testing, which are significantly more expensive and arduous regarding time and effort of retrieval. However, even when utilizing whole slide images, there lies challenges. Patient data can present very differently from case to case for a given disease, making analysis and survival predictions a difficult task. The advent of computer vision techniques and the success of deep learning offers a way to provide consistency in terms of decision making and leverage the ability to learn from a larger corpus of data than is possible for one person. The standardization between heterogenous patients and opportunity for discovery provided by machine learning offers a way to harness the full extent of information present in whole slide images. High performance and cloud computing has also made great strides to be able to support such computationally heavy workloads. The goal of this work is to tackle one complex medical task utilizing whole slide images and machine learning. Treatment decisions for ER+/HER2- patients are guided by a gold standard algorithm, the Oncotype DX Recurrence Score, that utilizes genetic assays to predict distance recurrence [25]. In this work, we explore the viability of a low-cost proxy by leveraging image-based features from whole slide images to provide the same prognostic and predictive information. We do so by identifying anomalous regions in whole slide images, extracting and exploring nuclear features to find discriminatory feature sets that maximize dissimilarity between two groups of Oncotype DX patients, and combining weak and strong recurrence related labels to create a robust deep learning classifier. We motivate a methodical approach that is supported by domain expertise to replace genetic features with image features to encourage a similar result as the Oncotype DX algorithm.
Raghu Machiraju (Advisor)
Zaibo Li (Committee Member)
Jeremy Morris (Committee Member)
58 p.
Computer Science
digital pathology, machine learning

Recommended Citations

Citations

  • Vroom, C. M. (2022). Machine Learning Models of Histopathologic Images to Serve as a Proxy to Predict Recurrence in ER+/HER- Breast Cancers [Master's thesis, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu1661139111525638

    APA Style (7th edition)

  • Vroom, Carolyn. Machine Learning Models of Histopathologic Images to Serve as a Proxy to Predict Recurrence in ER+/HER- Breast Cancers. 2022. Ohio State University, Master's thesis. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=osu1661139111525638.

    MLA Style (8th edition)

  • Vroom, Carolyn. "Machine Learning Models of Histopathologic Images to Serve as a Proxy to Predict Recurrence in ER+/HER- Breast Cancers." Master's thesis, Ohio State University, 2022. http://rave.ohiolink.edu/etdc/view?acc_num=osu1661139111525638

    Chicago Manual of Style (17th edition)

osu1661139111525638
38
© 2022, all rights reserved.
This open access ETD is published by The Ohio State University and OhioLINK.