Degree

Doctor of Philosophy (PhD)

Department

Division of Computer Science and Engineering

Document Type

Dissertation

Abstract

In Computer-Aided Diagnosis (CAD) of cancer, standard cost metrics (false-positives and false-negatives) fundamentally fail to account for overdiagnosis. Overdiagnosis is a critical scenario where a disease is correctly detected (true-positive) but is biologically indolent and would never have caused the patient harm or symptoms. While widely recognized in the medical community as a major healthcare crisis driving stressful and invasive overtreatment, overdiagnosis remains severely under-researched within computer science and engineering. This dissertation addresses this interdisciplinary gap by defining the three key computational challenges of overdiagnosis: (i) accurate estimation, (ii) harm quantification, and (iii) algorithmic mitigation. To overcome the estimation challenge, the unprecedented pause in mammography screening during the Covid-19 pandemic is leveraged as a natural experiment. This approach estimates the invasive breast cancer overdiagnosis rate at approximately 33.97%, revealing a strong inverse correlation between tumor size and indolence. To address harm quantification, Quality-Adjusted Life Expectancy (QALE) is utilized to mathematically measure the burden of overtreatment, demonstrating that overdiagnosed breast cancer patients lose an average of 24.455 Quality-Adjusted Life Days over a standard 15-month treatment cycle, and proving that CAD systems capable of detecting indolence yield massive population-level benefits. Finally, to address mitigation, this research develops a risk-adjusted data mining pipeline using an Extreme Gradient Boosting (XGBoost) classifier to safely distinguish malignant-aggressive from malignant-indolent trajectories. Evaluated individually across rigorous survival thresholds (5, 6, 7, 8, 9, and 10 years), the classifier employs a dynamic, asymmetric loss function on a 100-point scale. By enforcing a strict zero-false-positive boundary, the system successfully detects approximately 30% of indolent cases without ever misclassifying a lethal, aggressive cancer. Navigating the delicate ethical balance of minimizing breast cancer mortality while safely reducing overtreatment, this next-Generation CAD framework establishes a robust computational foundation for clinical therapeutic de-escalation.

Date

7-29-2026

Committee Chair

Triantaphyllou, Evangelos

LSU Acknowledgement

1

LSU Accessibility Acknowledgment

1

Download

Share

COinS