News & Events

Deep Learning for Medical Imaging: Automated Ultrasound Diagnosis and Physics-Informed Elastography Reconstruction

Donya Khaledyan

Supervised by Kevin Parker and Marvin Doyley

601 Computer Studies Building

This dissertation addresses key challenges in medical imaging analysis and reconstruction, with a focus on improving accessibility in ultrasound-based diagnosis and enhancing physical fidelity in optical coherence elastography (OCE). Despite significant advances in medical imaging, diagnostic accuracy remains limited by operator dependence and the difficulty of reconstructing reliable tissue properties from indirect measurements.

The first part of this work presents a multi-stage deep learning framework for the automated segmentation and classification of breast ultrasound volume sweep imaging (VSI) scans acquired by non-expert operators. The proposed approach integrates segmentation and classification into a unified pipeline, enabling reliable interpretation of ultrasound data without requiring specialized expertise. The framework is designed to be robust to noise and variability inherent in ultrasound imaging and demonstrates improved performance on both public and clinically acquired datasets, supporting scalable and accessible diagnostic workflows.

The second part focuses on image reconstruction in OCE, where the objective is to recover tissue mechanical properties from measured wavefield data. This problem is formulated as an image-to-image translation task. A physics-informed deep learning framework is introduced to incorporate domain-specific constraints into the reconstruction process, improving accuracy and stability. A transformer-enhanced architecture further captures long-range spatial dependencies, enabling more consistent and physically plausible reconstructions across experimental, phantom, and synthetic datasets.

Overall, this dissertation demonstrates that combining data-driven learning with physics-based modeling enables robust and generalizable solutions for both medical image analysis and reconstruction. These contributions advance the development of reliable, interpretable, and scalable imaging systems for real-world clinical applications.