Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

Early Disease Detection Through Computational Pathology

Burger, Virginia (2014) Early Disease Detection Through Computational Pathology. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

PDF (dissertation)
Primary Text

Download (46MB) | Preview


This thesis presents computational pathology algorithms for enabling early cancer detection in Barrett’s Esophagus (BE) and early subtype diagnosis in Interstitial Lung Diseases (ILD). BE is a condition affecting 10% of heartburn sufferers, for which 0.1% of patients develop esophageal adenocarcinoma each year. For most of the 130-200 diseases included in the class of ILDs, a full recovery is expected, but for a few of these diseases, the survival rate is less than three years. For both disease classes, treatment of the malignant forms would be harmful in patients with other forms, thus diagnosis is necessary prior to beginning treatment, and early treatment is most effective in eradicating disease. Early diagnosis of both of these disease classes is complicated by a high degree of sharing of subtle disease phenotypes, leading to high pathologist disagreement rates. Computational pathology methods can aid early diagnosis of these diseases through unbiased, data-driven algorithms.

To detect precancerous changes in patients with BE, we develop an automated algorithm which identifies epithelial nuclei in biopsy samples on which nano-scale optical biomarkers, related to cancer risk, can be quantified. The automated nuclei detector produces a higher quality selection of epithelial nuclei than manual detection, resulting in enhanced characterization of precancerous phenotype perturbations. To stratify ILD patients, we develop a novel quantitative representation of pathohistology samples that models lung architecture based on computed image features and insights from pathologists, and establish its utility as part of a diagnostic classifier. Algorithms such as these applied in a clinical setting can save pathologists time by filtering out obvious cases and providing unbiased reasoning to assist diagnoses.


Social Networking:
Share |


Item Type: University of Pittsburgh ETD
Status: Unpublished
CreatorsEmailPitt UsernameORCID
Burger, Virginiaginnyburger@gmail.comVMB12
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Thesis AdvisorChennubhotla, Chakrachakracs@pitt.eduCHAKRACS
Committee ChairTaylor, D Lansingdltaylor@pitt.eduDLTAYLOR
Committee MemberBenos, Panayiotis V.benos@pitt.eduBENOS
Committee MemberLiu, Yangliuy@pitt.eduLIUY
Committee MemberMiller,
Date: 14 January 2014
Date Type: Publication
Defense Date: 31 October 2013
Approval Date: 14 January 2014
Submission Date: 12 December 2013
Access Restriction: 1 year -- Restrict access to University of Pittsburgh for a period of 1 year.
Number of Pages: 140
Institution: University of Pittsburgh
Schools and Programs: School of Medicine > Computational Biology
Degree: PhD - Doctor of Philosophy
Thesis Type: Doctoral Dissertation
Refereed: Yes
Uncontrolled Keywords: computational pathology, Barrett's Esophagus, interstitial lung disease, holistic scene understanding, epithelial nuclei classification, nuclei segmentation
Date Deposited: 14 Jan 2015 06:00
Last Modified: 19 Dec 2016 14:41


Monthly Views for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item