📝 Publications

MORI-Seg: Learning Morphological Geometry for Instance Segmentation without Instance Annotations
Leiyue Zhao, Tianyu Shi, Daniel Reisenbuchler, Xinzi He, Junchao Zhu, Tianyuan Yao, Yuechen Yang, Yanfan Zhu, Junlin Guo, Gelei Xu, Haichun Yang, Yuankai Huo, Mert R. Sabuncu, Yihe Yang, Ruining Deng

• We propose MORI-Seg, a deep learning framework that enables instance segmentation of kidney functional units directly from semantic masks, without requiring instance-level annotations.
• By jointly modeling object-centric distance fields and boundary-band representations, MORI-Seg achieves improved instance separation and more reliable morphometric quantification compared with classical post-processing pipelines.

M³‑GloDets: multi-region and multi-scale analysis of fine‑grained diseased glomerular detection
Tianyu Shi, Xinzi He, Hongjin Fang, Kenji Ikemura, Mert R. Sabuncu, Yihe Yang, Ruining Deng

• We investigate multi-region and multi-scale strategies for fine-grained diseased glomerular detection in whole-slide renal pathology images.
• Our study systematically evaluates the impact of imaging magnification, patch size, and model design on multi-class glomerular analysis.

AI-Based Identification and Quantification of Global Glomerulosclerosis in Nephrectomy Specimens: Accuracy and Its Predictive Value
Ruining Deng, Tianyu Shi, Steven Salvatore, Surya Seshan, Brian Robinson, Mert Sabuncu, Yihe Yang

• We investigate AI-based quantification of global glomerulosclerosis from whole-slide renal pathology images using automated segmentation models.
• Our study compares AI-derived and pathologist-estimated glomerulosclerosis measurements and explores their associations with longitudinal renal function decline.

CloneShield: A Framework for Universal Perturbation Against Zero-Shot Voice Cloning
Renyuan Li, Zhibo Liang, Haichuan Zhang, Tianyu Shi, Zhiyuan Cheng, Jia Shi, Carl Yang, Mingjie Tang

Project

• We introduce CloneShield, a universal defense framework against zero-shot voice cloning.
• By generating imperceptible adversarial perturbations in the audio domain, our method protects speaker identity while maintaining natural audio quality, and effectively disrupts cloned speech generation across multiple state-of-the-art TTS systems.