Multi-Frame Super-Resolution for Fluoroscopic Imaging (Python & CUDA)

---

---

OXOS Medical is a leader in portable X-ray and fluoroscopic imaging technology, committed to improving real-time diagnostic imaging for orthopedic and interventional procedures. We seek an experienced image processing engineer to develop a Multi-Frame Adaptive Super-Resolution (MASR) approach incorporating Bayesian Super-Resolution techniques to enhance fluoroscopic images. The project will begin with a Python-based proof of concept (PoC) and later be optimized for real-time execution on an NVIDIA Jetson Xavier using CUDA. Project Overview Fluoroscopic X-ray imaging operates at low dose and high frame rates, leading to noisy, low-resolution images with motion artifacts. To improve image clarity while maintaining real-time performance, this project will implement a Multi-Frame Adaptive Super-Resolution (MASR) algorithm, leveraging Bayesian inference and motion compensation to reconstruct higher-quality images from fluoroscopy sequences. The ideal solution should: - Extract high-frequency details across multiple frames Reduce noise while preventing motion blur - Improve spatial resolution while preserving clinical details - Be optimized for real-time processing on NVIDIA Jetson Xavier Technical Scope The project involves developing an image processing pipeline with the following core components: - Multi-Frame Super-Resolution (MASR): Align and combine multiple consecutive fluoroscopic frames to enhance spatial resolution - Bayesian Super-Resolution (BSR): Probabilistically infer high-resolution image details using Bayesian inference models - Noise Reduction & Motion Compensation: Implement optical flow-based motion estimation to stabilize images before super-resolution - Contrast Enhancement & Edge Preservation: Apply CLAHE, adaptive histogram equalization, and edge-aware filters to improve visibility of anatomical structures and devices - Real-Time Optimization: Deploy and optimize the final pipeline on NVIDIA Jetson Xavier, leveraging CUDA, TensorRT, and VPI Project Inputs & Outputs - Inputs: Fluoroscopic image sequences (grayscale images, DICOM format or PNG sequences) - Outputs: High-resolution, noise-reduced radiographic images suitable for medical diagnosis Definition of Success A successful implementation will: - Deliver a Python-based PoC demonstrating MASR with Bayesian inference on fluoroscopic images - Implement real-time motion compensation and multi-frame fusion for fluoroscopic sequences - Optimize the final solution for Jetson Xavier using CUDA, achieving at least 15 FPS real-time processing - Provide well-documented, modular, and reusable code with performance benchmarks and validation on real fluoroscopic data Ideal Experience - Strong background in image processing and super-resolution algorithms - Experience with medical imaging (X-ray, fluoroscopy, CT, etc.) - Proficiency in Python (NumPy, SciPy, OpenCV, PyTorch, TensorFlow) - Experience with Bayesian inference methods for image processing - Expertise in CUDA, NVIDIA Jetson, TensorRT, and real-time GPU optimization - Familiarity with optical flow techniques for motion estimation

Keyword: Python Development

Price: $50.0

Python OpenCV Image Processing Medical Imaging