Congratulations to Dr. Xuzhe Zhang!!!

Abstract

Robust Artificial Intelligence for Medical Image Analysis

by Xuzhe Zhang

Medical imaging is indispensable to modern healthcare, but the growing heterogeneities and developments of scanners, protocols, and techniques expose AI systems to an evolving data distribution, degrading the performance of models trained on previous data and hindering large-scale deployment of AI in medical image analysis. This thesis focuses on addressing this technical gap through self- and semi-supervised frameworks that learn from heterogeneous and newly acquired data without requiring additional manual annotations. These methods enable segmentation and quantification of image-based biomarkers that remain robust across study centers, imaging sequences, protocols, and even generations of imaging modalities. The overarching goal is to develop robust AI models that can be applied to real-world, heterogeneous, and continually evolving data distributions.

We first propose a unified domain-adaptive segmentation framework named MAPSeg and conduct the first systematic evaluation of an unsupervised domain-adaptation framework under various domain shifts and data-access scenarios. MAPSeg is designed for practical deployment: it supports centralized training when data can be collected, privacy-preserving federated learning when data cannot move, and test-time adaptation when only unlabeled target data are available. It demonstrates superior robustness against common domain shifts in the neonatal infant-brain segmentation task.

Building on this foundation, we develop the first fully automatic segmentation pipeline to delineate pulmonary airway trees from MRI—a task previously considered impossible until recent advances in MRI acquisition sequences and reconstruction algorithms—without using any MRI annotations. The resulting framework enables radiation-free quantification of pulmonary airways and paves the way for repeated scans and longitudinal monitoring of the mismatch between airway caliber and lung size in pediatric populations, a phenomenon believed to arise early in life and linked to increased risk of Chronic Obstructive Pulmonary Disease (COPD) in later years.

Finally, the work introduces MDLEmph, an acquisition-agnostic, multimodal framework for emphysema quantification on chest CT. MDLEmph fuses image features with protocol-aware priors and scanner metadata and trains with distribution augmentations to learn protocol-invariant representations. This results in harmonized emphysema assessments that track patient trajectories consistently across centers and years—strengthening retrospective analyses, enabling fair pooling of heterogeneous cohorts, and improving the interpretability of imaging endpoints in both observational studies and clinical trials.

Together, these contributions provide a principled and scalable path toward robust and trustworthy AI-assisted medical image analysis under domain shift. By unifying unsupervised adaptation, scalable self- and semi-supervised learning, and multimodal integration, this thesis advances AI-driven computational tools that broaden the reach of neurodevelopmental research, enable longitudinal quantitative airway studies from early life, and deliver reproducible biomarkers for chronic respiratory disease—ultimately helping translate AI from benchmarks to real-world medical research and clinical practice.