Date of Award

January 2023

Document Type

Open Access Thesis

Degree Name

Master of Public Health (MPH)


School of Public Health

First Advisor

Shuangge Ma

Second Advisor

Yuan Huang


Objectives. To assess the reproducibility and overlapping of information in the automated pathological imaging features pipeline for lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC) and skin cutaneous melanoma (SKCM); and to construct an integrated prognostic model using these pathological imaging features for LUAD, LUSC and SKCM. Methods. 470 LUAD patients, 477 LUSC patients and 428 SKCM patients are included in the study. Whole-slide histopathology images as well as demographical and clinical data are obtained from the TCGA website. Histopathological images are processed using the Openslide Python library and chopped into twenty small 500*500- or 700*700-pixel subimages. CellProfiler software was used to extract imaging features and provide its statistical values. RV coefficients are computed to assess robustness and reproducibility of the pipeline. Cox Proportional Hazards Model with Principal Component Analysis, and Cox Proportional Hazards Model with elastic net regularization are used to construct the survival model. Concordance indexes are calculated to measure the predictivity of the models using random splitting-based evaluation.Results. The high RV coefficients indicate the robustness and reliability of the imaging feature pipeline in the context of extracting feature statistical values. By using the random splitting- based evaluation, we observe that pathological imaging data has moderate/weak predictive performance for LUAD, LUSC and SKCM, respectively. Conclusion. This study suggests that histopathological imaging features extracted from CellProfiler software may not be predictive for LUAD, LUSC and SKCM prognosis, but provide some broadly insight into cancer prognosis modeling and analysis. Future large-scale studies that incorporate other cell properties analysis software are needed.


This is an Open Access Thesis.

Open Access

This Article is Open Access