Real-Time Kidney-Stone Detection
Kidney-stone material has to be spotted and localized in continuous microscopic-camera video as it streams — a real-time detection problem, not an offline image-classification one.
Ongoing academic computer vision research at NYU Tandon's Flexible AI-Enabled Mechatronic Systems (FAMS) Lab. The current work develops a real-time pipeline that detects and localizes kidney stones in microscopic-camera video: preparing and preprocessing image/video data, developing and evaluating detection models, and running inference on a live stream. Earlier work in the lab covered endoscopic image analysis with YOLOv5 object detection, U-Net segmentation, and visualization in 3D Slicer. Involvement began through NYU's Vertically Integrated Projects program in Fall 2025; Volunteer Research Assistant since September 2026.
Current focus
- Image/video preprocessing for microscopic-camera footage
- Kidney-stone detection and localization
- Model development and evaluation
- Live-stream inference
Current status
Ongoing academic research at NYU's FAMS Lab — not a deployed system. No accuracy, FPS, latency, dataset-size, or model benchmarks are claimed here. This page will grow as real results become available; until then, it documents the current focus — the pipeline figure above is conceptual, not the lab's setup — not a finished capability.
Stack
PyTorch / OpenCV / YOLOv5 / U-Net / 3D Slicer