Intelligent Analysis of Medical Images Based on Improved U-Net and SIFT Algorithms for Pre-Hospital Emergency Care

Background/Objectives: Timely and accurate medical imaging in pre-hospital emergency care is crucial for improving the effectiveness of emergency medical rescue. Methods: In this study, a portable endoscopic system was developed for image acquisition and intelligent analysis. The core innovation lies in a dedicated image processing framework that integrates an improved U-Net neural network for real-time image dehazing and an enhanced Scale-Invariant Feature Transform (SIFT) algorithm for precise image stitching. Results: Experimental results demonstrated the exceptional performance of our method: the dehazing U-Net achieved a Structural Similarity Index (SSIM) of 0.98, a Peak Signal-to-Noise Ratio (PSNR) of 31.05, and a processing speed exceeding 70 frames per second (fps), while the enhanced SIFT algorithm effectively minimized stitching artifacts and vessel misalignment, yielding an SSIM of 0.9367 and a PSNR of 35.9768. Conclusions: This system significantly enhanced image quality and processing speed, enabling the acquisition of precise imaging information to support rapid diagnosis at the emergency scene. The findings establish a solid foundation for advancing pre-hospital emergency medical imaging and suggest promising avenues for future validation with diverse datasets and long-term clinical evaluations to further improve algorithmic robustness and applicability.