Puja Saha

Adaptive differential privacy for federated medical segmentation across modalities and complexity levels

Monday, February 16, 2026 · 5:30 – 7:00 PM PST
East Exhibition Hall A (Level 1)

Pramit Dutta

Structured clinical interpretation of lung computed tomography: a post‑detection application of vision language models

Tuesday, February 17, 2026 · 2:20 – 2:40 PM PST
East 15 (Level 2)

Zachary Szentimrey

Development of a deep learning liver segmentation model for patients with liver disease imaged with handheld point of care ultrasound

Tuesday, February 17, 2026 · 2:10 – 2:30 PM PST
East 8 (Level 2)

Zinah Ghulam

Severity‑aware multimodal network for chest x‑ray triage and disease captioning

Monday, February 16, 2026 · 5:30 – 7:00 PM PST
East Exhibition Hall A (Level 1)

Zinah's Poster for SPIE Medical Imaging 2026

🎓 Puja Saha – MASc Graduate

Thesis: Optimizing Differentially Private Federated Learning for Medical Image Segmentation

Puja's thesis focused on developing a privacy-preserving and scalable federated learning framework for medical image segmentation, with methods specifically designed to minimize the privacy-induced trade-off in model performance.

This approach can enable the use of vast amounts of medical data across institutions that often remain under-utilized, to develop robust medical AI models, all while adhering to privacy regulations such as HIPAA, PIPEDA, and GDPR. She evaluated the approach across diverse imaging modalities (2D and 3D), varying levels of complexity, and real-world institutional heterogeneity, demonstrating significantly improved performance compared to conventional differential privacy-based federated learning methods.

Puja will soon be presenting this work at SPIE Medical Imaging. A journal article is also forthcoming, covering the method's effectiveness under real-world institutional heterogeneity.

🎓 Helena Kunic – MASc Graduate

Thesis: Ultrasound Based Evaluation of Knob Modifications in Pessaries for Stress Urinary Incontinence

Defense Date: Wednesday, August 27, 2025, 9:00 AM

Helena's research investigated the effects of modifying the knob angle and size of ring pessaries in treating stress urinary incontinence (SUI). Through in-depth analysis using clinical trials, patient feedback, and ultrasound imaging, she demonstrated improved symptom relief in some women. Additionally, Helena developed a machine learning model to automatically segment key pelvic structures, advancing objective assessment of treatment outcomes.

🎓 Negin Piran Nanekaran – PhD Graduate

Thesis: Deep Learning and Federated Learning for Prostate Cancer Classification and Recurrence Prediction

Defense Date: Thursday, September 4, 2025, 1:30 PM

Negin addressed key challenges in medical imaging and predictive oncology. Her dissertation developed a deep learning framework for predicting 5-year biochemical recurrence in prostate cancer patients and introduced a novel federated learning approach for multi-institutional MRI data analysis. Her methods demonstrated superior prediction accuracy, privacy-preserving efficiency, and potential for broad clinical application.