Near-infrared image-based vascular network extraction using enhanced matched filtering

Intravenous injection and venipuncture are vital in modern medicine, making the positioning of surface veins crucial. Utilizing the absorption characteristics of deoxyhemoglobin in the near-infrared spectrum, a vascular network recognition algorithm using near-infrared images and matched filtering is proposed. An 850 nm LED serves as the auxiliary lighting source, and the Contrast Limited Adaptive Histogram Equalization (CLAHE) method improves infrared image contrast. Based on the morphological characteristics of the infrared vein images, enhanced matched filtering successfully extracts vessels, increasing the signal-to-noise ratio and enabling accurate extraction of even small vessels. The algorithm’s reliability is validated with the Digital Retinal Images for Vessel Extraction (DRIVE) dataset, achieving a Dice coefficient of 73.92. Due to its low computational complexity, fast speed, and excellent performance, a handheld portable vein projector has been developed, featuring real-time infrared image acquisition, processing, and in-situ vein distribution projection, and successfully applied in medical practice.